MBD-PT 2023 Social Network Analysis

---

Analysing Drug Interactions and Chemical Compound Similarities for Business Insights - Graph Metrics

Table of contents

• 0. Description & General Objective
• 1. Import Libraries
• 2. Import Datasets
  • 2.1. DDINTER
  • 2.2. ATC
  • 2.3. IQVIA
• 3. Create Nodes Dataset
  • 3.1. Drug Name Replacement
  • 3.2. Add Molecule Characteristics from CHEMBL
  • 3.3. Add Chembl Database ATC Classifications
    • 3.3.1. ATC Code Classifications
    • 3.3.2. Adding ATC1 code to the table
    • 3.3.3 Adding ATC2 code to the table
• 4. Graph Statistics
  • 4.1) Graph Summary
  • 4.2) ATC Plots and Distribution
• 5. Graph Basic Metrics
  • 5.1) Closeness centrality
  • 5.2) Betweeness centrality
  • 5.3) Pagerank
  • 5.4) Degree centrality
  • 5.5) Eigenvector centrality
  • 5.6) Cluster Coefficient
• 6.Drug Graph Metrics
  • 6.1) Transitivity
  • 6.2) K-Core Decomposition
  • 6.3) Dispersion
  • 6.4) Assortativity Coefficient
  • 6.5) Jaccard similarity coefficient
  • 6.6) Link Prediction

0.Descriptive and General Objective

Goal: To identify potential drug interactions and similarities based on both social network graphs and chemical structures.

Objective: To improve patient safety and optimize drug therapy by providing healthcare professionals with a tool that can quickly and accurately analyze potential drug interactions and similarities.

1. Import Libraries

import os
import pandas as pd

import pandas as pd
import networkx as nx
import matplotlib.pyplot as plt

import plotly.graph_objs as go

import re

import operator

import time


from chembl_webresource_client.new_client import new_client

2. Import Datasets

2.1 DDINTER

Import Ddinter into a Pandas Dataframe

Drug-drug interaction (DDI) is broadly described as a change in the effect of one drug due to the presence of another drug. Most interactions can seriously affect efficacy and safety drug profiles and have been a significant cause of adverse drug interactions (ADRs). DDIs can be divided into pharmaceutical, pharmacokinetic (PK) interactions, and pharmacodynamic (PD) interactions. Pharmaceutical interactions occur due to physical or chemical incompatibilities, which are often out of the conventional scope of DDIs. PK interactions could influence the drug disposition process to change the free-drug concentration at the target site, including absorption, distribution, metabolism, and excretion (ADME). While PD interactions usually are synergistic or antagonistic, which can occur by interacting with the same therapeutic targets, involving in different signaling pathways, or causing different pharmacological responses. We mainly focus on the latter two kinds of DDIs in this work and the details of these subclasses are elaborated

Criteria for severity annotation Categories of severity were accepted as suggested by DRUGDEX and other similar resources:

Major: The interactions are life-threatening and/or require medical treatment or intervention to minimize or prevent severe adverse effects.

Moderate:The interactions may result in exacerbation of the disease of the patient and/or change in therapy.

Minor: The interactions would limit the clinical effects. The manifestations may include an increase in frequency or severity of adverse effects, but usually they do not require changes in therapy.

Unknown: The DDIs collected from the article published in Sci Transl Med were lack of mechanism descriptions, and thus the severity classifications of these DDIs were annotated with 'Unknown'.

# directory containing the csv files
directory = '/Users/stephaniegessler/Documents/Social Network Analysis/ddinter'

# list of csv files in the directory
csv_files = [f for f in os.listdir(directory) if f.endswith('.csv')]

# empty list to store dataframes
dfs = []

# loop through csv files and create dataframes
for csv_file in csv_files:
    # read csv file
    df = pd.read_csv(os.path.join(directory, csv_file))
    
    
    # append dataframe to the list
    dfs.append(df)

# concatenate dataframes into one
combined_df = pd.concat(dfs, ignore_index=True)

#remove duplicates of the entire row 
#combined_df.drop_duplicates(subset=None, keep='first', inplace=True)

# save combined dataframe as a csv file
combined_df.to_csv('combined_csv_files.csv', index=False)

# Remove duplicates of the entire row
combined_df.drop_duplicates(subset=None, keep='first', inplace=True)

combined_df.shape

(160235, 5)

# get unique values in the 'Level' column
unique_levels = combined_df['Level'].unique()

# print unique values
print(unique_levels)

['Minor' 'Moderate' 'Major' 'Unknown']

# define a lambda function to map level to weight
map_weight = lambda x: 2 if x == 'Unknown' else 3 if x == 'Minor' else 2 if x == 'Moderate' else 1 if x == 'Major' else 0

# add new column 'Weight' with mapped values
combined_df['Weight'] = combined_df['Level'].apply(map_weight)

# print dataframe with new column
combined_df

	DDInterID_A	Drug_A	DDInterID_B	Drug_B	Level	Weight
0	DDInter513	Dexamethasone	DDInter582	Dolutegravir	Minor	3
1	DDInter997	Ivacaftor	DDInter582	Dolutegravir	Minor	3
2	DDInter1019	Lamivudine	DDInter1123	Mannitol	Moderate	2
3	DDInter195	Betamethasone	DDInter1961	Zidovudine	Moderate	2
4	DDInter1961	Zidovudine	DDInter246	Budesonide	Moderate	2
...	...	...	...	...	...	...
222341	DDInter856	Heparin	DDInter805	gamma-Aminobutyric acid	Unknown	2
222342	DDInter1683	Sodium bicarbonate	DDInter805	gamma-Aminobutyric acid	Unknown	2
222359	DDInter908	Iloprost	DDInter62	Ambrisentan	Unknown	2
222365	DDInter1951	Warfarin	DDInter1942	Vitamin D	Unknown	2
222368	DDInter1683	Sodium bicarbonate	DDInter1942	Vitamin D	Unknown	2

160235 rows × 6 columns

#Create edges table from combined_df
edges = combined_df

#replace "," with ";" 
edges['Drug_A'] = edges['Drug_A'].str.replace(';', ',')
edges['Drug_B'] = edges['Drug_B'].str.replace(';', ',')

#convert combined_df to csv file with ";" as separator
edges.to_csv('edges.csv', sep=';', index=False)

2.2 Importing ATC Table

The Anatomical Therapeutic Chemical (ATC) classification will be imported to the dataset. ATC will be explained under the ATC section.

ATC = pd.read_excel('/Users/stephaniegessler/IE Students/Term 3 - Data Dynamos - General/Social Network Analysis/drugs/swissmedic/ATC Classifications and Names.xlsx')

ATC.head(20)

	atc_code	atc_name	ddd	uom	adm_r	note	Unnamed: 6
0	A	ALIMENTARY TRACT AND METABOLISM	NaN	NaN	NaN	NaN	ATC1
1	A01	STOMATOLOGICAL PREPARATIONS	NaN	NaN	NaN	NaN	ATC2
2	A01A	STOMATOLOGICAL PREPARATIONS	NaN	NaN	NaN	NaN	ATC3
3	A01AA	Caries prophylactic agents	NaN	NaN	NaN	NaN	ATC4
4	A01AA01	sodium fluoride	2023-01-01 00:00:00	mg	O	0.5 mg fluoride	ATC5
5	A01AA02	sodium monofluorophosphate	NaN	NaN	NaN	NaN	NaN
6	A01AA03	olaflur	2023-01-01 00:00:00	mg	O	NaN	NaN
7	A01AA04	stannous fluoride	NaN	NaN	NaN	NaN	NaN
8	A01AA30	combinations	NaN	NaN	NaN	NaN	NaN
9	A01AA51	sodium fluoride, combinations	NaN	NaN	NaN	NaN	NaN
10	A01AB	Antiinfectives and antiseptics for local oral ...	NaN	NaN	NaN	NaN	NaN
11	A01AB02	hydrogen peroxide	60	mg	O	NaN	NaN
12	A01AB03	chlorhexidine	30	mg	O	NaN	NaN
13	A01AB04	amphotericin B	40	mg	O	NaN	NaN
14	A01AB05	polynoxylin	0.18	g	O	NaN	NaN
15	A01AB06	domiphen	3	mg	O	NaN	NaN
16	A01AB07	oxyquinoline	80	mg	O	NaN	NaN
17	A01AB08	neomycin	NaN	NaN	NaN	NaN	NaN
18	A01AB09	miconazole	0.2	g	O	NaN	NaN
19	A01AB10	natamycin	20	mg	O	NaN	NaN

2.3 Import IQVIA

Import IQVIA into a Pandas Dataframe which provides the units distribution of the market in Switzerland by the different molecule/ drug type. Some drugs are used in singular in a mono therapy, others are put in combination as combi. The ranking describes the distribution of these molecules/Drugs based on 2023 results and usage in the market. Units MAT (Moving Annual Total) provides the market usage and distribution in the market. Our analysis will focus only on the current market 2023. The Sum inlc Combi provides the distribution of Drugs which are given in Molecule with the molecule mentioned. The Ranking Mono and Ranking Combi provides the information which of these are ranked the highest in terms of distriburion only looking at the Mono Drug or in Combination with others. For similicity, we do not consider packing size and strength of the respective molecule/drug.

#rettrive from excel the second workbook "Mono"
IQVIA = pd.read_excel('/Users/stephaniegessler/IE Students/Term 3 - Data Dynamos - General/Social Network Analysis/drugs/CH Drug Distribution/IQVIA.xlsx',sheet_name="Mono")

IQVIA.head(20)

	Row Labels	Units MAT Feb 2020	Units MAT Feb 2021	Units MAT Feb 2022	Units MAT Feb 2023	% Growth	Sum incl combi	Ranking Mono	Ranking Combi
0	PARACETAMOL	8709710.0	8605565.0	9758040.0	11047056.0	0.132098	14291456	1	1
1	IBUPROFEN	8022915.0	6257911.0	7808308.0	9724304.0	0.245379	9850860	2	2
2	COLECALCIFEROL	3928582.0	4714102.0	4769101.0	4308630.0	-0.096553	5929035	3	4
3	DICLOFENAC	4434500.0	4068832.0	4224026.0	4231062.0	0.001666	5125490	4	5
4	PANTOPRAZOLE	2899125.0	2879570.0	3037873.0	3170851.0	0.043773	3170851	5	11
5	METAMIZOLE SODIUM	2333573.0	2304660.0	2359320.0	2316965.0	-0.017952	2316965	6	19
6	ACETYLSALICYLIC ACID	2323379.0	2279675.0	2277186.0	2274078.0	-0.001365	2812337	7	15
7	MAGNESIUM	2046801.0	2067960.0	2154563.0	2185350.0	0.014289	3240089	8	10
8	XYLOMETAZOLINE	1904461.0	1376550.0	1658246.0	1999905.0	0.206036	3937333	9	9
9	ACETYLCYSTEINE	1536567.0	900140.0	1116398.0	1834454.0	0.643190	2500285	10	17
10	ZOLPIDEM	1439833.0	1389898.0	1365755.0	1552515.0	0.136745	1552515	11	30
11	DEXPANTHENOL	1461137.0	1412184.0	1454316.0	1511731.0	0.039479	4260796	12	7
12	METFORMIN	1247791.0	1264797.0	1290373.0	1311408.0	0.016301	1826206	13	24
13	ATORVASTATIN	1144657.0	1165781.0	1209738.0	1252593.0	0.035425	1497215	14	34
14	LEVOTHYROXINE SODIUM	1179684.0	1168192.0	1192707.0	1239288.0	0.039055	1252416	15	40
15	LORAZEPAM	1133911.0	1121817.0	1109016.0	1148151.0	0.035288	1148154	16	45
16	MOMETASONE	955500.0	864249.0	921392.0	1088789.0	0.181678	1098495	17	52
17	UREA	1068512.0	1061007.0	1049135.0	996507.0	-0.050163	2575140	18	16
18	LOPERAMIDE	992700.0	624831.0	798828.0	986729.0	0.235221	995494	19	57
19	CETIRIZINE	821062.0	975891.0	874538.0	968601.0	0.107557	1284112	20	39

3. Create Nodes Dataset

We are creating a Nodes Dataset where we add all additional attributs.

#create a nodes tables
# Creata a new pandas datframe for the nodes characteristics, use DDInterID_A and DDInterID_B and Drug_A and DrugB as label. Make sure they are unique and drop duplicates
nodes = pd.DataFrame(combined_df[['DDInterID_A', 'Drug_A']].drop_duplicates().values, columns=['DDInterID', 'Drug'])
nodes = nodes.append(pd.DataFrame(combined_df[['DDInterID_B', 'Drug_B']].drop_duplicates().values, columns=['DDInterID', 'Drug']), ignore_index=True)
nodes = nodes.drop_duplicates()

/var/folders/ks/5bc1x9p158vgc4774v7r2tq40000gn/T/ipykernel_65890/2044511243.py:4: FutureWarning:

The frame.append method is deprecated and will be removed from pandas in a future version. Use pandas.concat instead.

nodes.head()

	DDInterID	Drug
0	DDInter513	Dexamethasone
1	DDInter997	Ivacaftor
2	DDInter1019	Lamivudine
3	DDInter195	Betamethasone
4	DDInter1961	Zidovudine

#count the number of nodes
nodes.shape

(1939, 2)

3.1. Drug Name Replacement

In order to harmonize Chembl, IQVIA and the DDInter Database, we harmonize the database Drug names to retrieve the right information under the name.

#Dublicate the nodes['Drug'] to a new column Drug_original
nodes['Drug_original'] = nodes['Drug']

# replace Insulin human (isophane) in Drug and Drug_B with INSULIN HUMAN ISOPHANE
nodes['Drug'] = nodes['Drug'].replace('Insulin human (isophane)', 'INSULIN HUMAN ISOPHANE')

# replace Insulin human (regular) in Drug and Drug_B with INSULIN HUMAN BASE
nodes['Drug'] = nodes['Drug'].replace('Insulin human (regular)', 'INSULIN HUMAN BASE')

# replace Insulin human (zinc) in Drug and Drug_B with 	INSULIN HUMAN ZINC
nodes['Drug'] = nodes['Drug'].replace('Insulin human (zinc)', 'INSULIN HUMAN ZINC')

# replace Insulin human (zinc extended) in Drug and Drug_B with INSULIN HUMAN ZINC, EXTENDED
nodes['Drug'] = nodes['Drug'].replace('Insulin human (zinc extended)', 'INSULIN HUMAN ZINC, EXTENDED')

# replace Dextran (-1) in Drug and Drug_B with 	DEXTRAN 1
nodes['Drug'] = nodes['Drug'].replace('Dextran (-1)', 'DEXTRAN 1')

# replace Dextran (high molecular weight) in Drug and Drug_B with 	DEXTRAN, HIGH MOLECULAR WEIGHT
nodes['Drug'] = nodes['Drug'].replace('Dextran (high molecular weight)', 'DEXTRAN, HIGH MOLECULAR WEIGHT')

# replace Amphotericin B (lipid complex) in Drug_A and Drug_B with AMPHOTERICIN B
nodes['Drug'] = nodes['Drug'].replace('Amphotericin B (lipid complex)', 'AMPHOTERICIN B')

# replace Cocaine (nasal) in Drug_A and Drug_B COCAINE HYDROCHLORIDE
#nodes['Drug'] = nodes['Drug'].replace('Cocaine (nasal)', 'COCAINE HYDROCHLORIDE')

# replace Daunorubicin (liposomal) in Drug_A and Drug_B with DAUNORUBICIN CITRATE
nodes['Drug'] = nodes['Drug'].replace('Daunorubicin (liposomal)', 'DAUNORUBICIN CITRATE')

# replace DDoxorubicin (liposomal) in Drug_A and Drug_B with DOXORUBICIN HYDROCHLORIDE
nodes['Drug'] = nodes['Drug'].replace('Doxorubicin (liposomal)', 'DOXORUBICIN HYDROCHLORIDE')

# replace Amphotericin B (liposomal) in Drug_A and Drug_B with AMPHOTERICIN B LIPOSOME
nodes['Drug'] = nodes['Drug'].replace('Amphotericin B (liposomal)', 'AMPHOTERICIN B LIPOSOME')

# replace Benzylpenicillin (sodium) in Drug_A and Drug_B with PENICILLIN G
nodes['Drug'] = nodes['Drug'].replace('Benzylpenicillin (sodium)', 'PENICILLIN G SODIUM')

# replace Heparin (flush) in Drug_A and Drug_B with HEPARIN SODIUM
nodes['Drug'] = nodes['Drug'].replace('Heparin (flush)', 'HEPARIN')

# replace Salicylic acid (sodium) in Drug_A and Drug_B with SODIUM SALICYLATE
nodes['Drug'] = nodes['Drug'].replace('Salicylic acid (sodium)', 'SODIUM SALICYLATE')

# replace Factor IX Complex (Human) in Drug_A and Drug_B with FACTOR IX COMPLEX HUMAN
nodes['Drug'] = nodes['Drug'].replace('Factor IX Complex (Human)', 'FACTOR IX COMPLEX HUMAN')

# replace Polyethylene glycol (3350 with electrolytes) in Drug_A and Drug_B with POLYETHYLENE GLYCOL 3350
nodes['Drug'] = nodes['Drug'].replace('Polyethylene glycol (3350 with electrolytes)', 'POLYETHYLENE GLYCOL 3350')


# replace Polyethylene glycol (3350) in Drug_A and Drug_B with POLYETHYLENE GLYCOL
nodes['Drug'] = nodes['Drug'].replace('Polyethylene glycol (3350)', 'POLYETHYLENE GLYCOL 3350')

# replace Typhoid vaccine (live) in Drug_A and Drug_B with TYPHOID VACCINE
nodes['Drug'] = nodes['Drug'].replace('Typhoid vaccine (live)', 'TYPHOID VACCINE')

# replace Human Rho(D) immune globulin in Drug_A and Drug_B with RHO (D) IMMUNE GLOBULIN
nodes['Drug'] = nodes['Drug'].replace('Human Rho(D) immune globulin', 'RHO (D) IMMUNE GLOBULIN')


# replace Fludeoxyglucose (18F) in Drug_A and Drug_B with FLUDEOXYGLUCOSE F 18
nodes['Drug'] = nodes['Drug'].replace('Fludeoxyglucose (18F)', 'FLUDEOXYGLUCOSE F 18')

# replace Olopatadine (nasal) in Drug_A and Drug_B with OLOPATADINE HYDROCHLORIDE
#nodes['Drug'] = nodes['Drug'].replace('Olopatadine (nasal)', 'OLOPATADINE HYDROCHLORIDE')

# replace Tositumomab (I-131) in Drug_A and Drug_B with TOSITUMOMAB 131I
nodes['Drug'] = nodes['Drug'].replace('Tositumomab (I-131)', 'TOSITUMOMAB 131I')

# replace Human immunoglobulin G (intravenous) in Drug_A and Drug_B with HUMAN IMMUNOGLOBULIN G
nodes['Drug'] = nodes['Drug'].replace('Human immunoglobulin G (intravenous)', 'HUMAN IMMUNOGLOBULIN G')


# replace Antilymphocyte immunoglobulin (horse) in Drug_A and Drug_B with ANTILYMPHOCYTE IMMUNOGLOBULIN (HORSE)
nodes['Drug'] = nodes['Drug'].replace('Antilymphocyte immunoglobulin (horse)', 'ANTILYMPHOCYTE IMMUNOGLOBULIN (HORSE)')

# replace Vincristine (liposome) in Drug_A and Drug_B with VINCRISTINE SULFATE
nodes['Drug'] = nodes['Drug'].replace('Vincristine (liposome)', 'VINCRISTINE')


# replace Insulin aspart (aspart protamine) in Drug_A and Drug_B with INSULIN ASPART PROTAMINE RECOMBINANT
nodes['Drug'] = nodes['Drug'].replace('Insulin aspart (aspart protamine)', 'INSULIN ASPART PROTAMINE RECOMBINANT')

# replace Manganese (sulfate) in Drug_A and Drug_B with MANGANESE SULFATE
nodes['Drug'] = nodes['Drug'].replace('Manganese (sulfate)', 'MANGANESE SULFATE')

# replace Azelastine (nasal) in Drug_A and Drug_B with AZELASTINE
#nodes['Drug'] = nodes['Drug'].replace('Azelastine (nasal)', 'AZELASTINE')

# replace Insulin lispro (protamine) in Drug_A and Drug_B with 	INSULIN LISPRO PROTAMINE RECOMBINANT
nodes['Drug'] = nodes['Drug'].replace('Insulin lispro (protamine)', 'INSULIN LISPRO PROTAMINE RECOMBINANT')

# replace Prilocaine (topiclal) in Drug_A and Drug_B with PRILOCAINE
nodes['Drug'] = nodes['Drug'].replace('Prilocaine (topiclal)', 'Prilocaine (topical)')

# replace Iobenguane (I-131) in Drug_A and Drug_B with 	IOBENGUANE I 131
nodes['Drug'] = nodes['Drug'].replace('Iobenguane (I-131)', 'IOBENGUANE I 131')

# replace Iobenguane (I-123) in Drug_A and Drug_B with IOBENGUANE I 123
nodes['Drug'] = nodes['Drug'].replace('Iobenguane (I-123)', 'IOBENGUANE I 123')

# replace Sodium phosphate, monobasic (p32) in Drug_A and Drug_B SODIUM PHOSPHATE, MONOBASIC
nodes['Drug'] = nodes['Drug'].replace('Sodium phosphate, monobasic (p32)', 'SODIUM PHOSPHATE, MONOBASIC')

# replace Manganese (chloride) in Drug_A and Drug_B with MANGANESE CHLORIDE
nodes['Drug'] = nodes['Drug'].replace('Manganese (chloride)', 'MANGANESE CHLORIDE')

# replace Levomefolic acid (calcium) in Drug_A and Drug_B with LEVOMEFOLIC ACID
nodes['Drug'] = nodes['Drug'].replace('Levomefolic acid (calcium)', 'LEVOMEFOLIC ACID')

# replace Dextran (low molecular weight) in Drug with DEXTRAN
nodes['Drug'] = nodes['Drug'].replace('Dextran (low molecular weight)', 'DEXTRAN')

# replace Prilocaine (topiclal) in Drug with PRILOCAINE
nodes['Drug'] = nodes['Drug'].replace('Prilocaine (topiclal)', 'Prilocaine (topical)')

# replace Ranitidine (bismuth citrate) in Drug with RANITIDINE BISMUTH CITRATE
nodes['Drug'] = nodes['Drug'].replace('Ranitidine (bismuth citrate)', 'RANITIDINE BISMUTH CITRATE')

# replace Samarium (153Sm) lexidronam in Drug with SAMARIUM SM 153 LEXIDRONAM PENTASODIUM
nodes['Drug'] = nodes['Drug'].replace('Samarium (153Sm) lexidronam', 'SAMARIUM SM 153 LEXIDRONAM PENTASODIUM')

# replace Benzylpenicillin (potassium) in Drug with PENICILLIN G POTASSIUM
nodes['Drug'] = nodes['Drug'].replace('Benzylpenicillin (potassium)', 'PENICILLIN G POTASSIUM')

# replace Chorionic Gonadotropin (Human) in Drug with GONADOTROPIN, CHORIONIC
nodes['Drug'] = nodes['Drug'].replace('Chorionic Gonadotropin (Human)', 'GONADOTROPIN, CHORIONIC')

# replace Bupivacaine (liposome) in Drug with BUPIVACAINE
nodes['Drug'] = nodes['Drug'].replace('Bupivacaine (liposome)', 'BUPIVACAINE')

# replace 	Insulin aspart (aspart) in Drug with INSULIN ASPART
nodes['Drug'] = nodes['Drug'].replace('Insulin aspart (aspart)', 'INSULIN ASPART')

# replace Cytarabine (liposomal) in Drug with CYTARABINE
nodes['Drug'] = nodes['Drug'].replace('Cytarabine (liposomal)', 'CYTARABINE')



# remove from Drug_A and Drug_B the only  " (ophthalmic)" from all names of the drug
#nodes['Drug'] = nodes['Drug'].replace('(-ophthalmic)', '')

# remove from Drug_A and Drug_B the only  "(topical)" from all names of the drug
#nodes['Drug'] = nodes['Drug'].replace('(-(topical))', '')

# remove from Drug_A and Drug_B the only  " (ophthalmic)" from all names of the drug

#nodes['Drug'] = nodes['Drug'].apply(lambda x: re.sub(r'\( *ophthalmic *\)', '', x))

# remove from Drug_A and Drug_B the only  "(topical)" from all names of the drug


#nodes['Drug'] = nodes['Drug'].apply(lambda x: re.sub(r'\( *topical *\)', '', x))

#nodes['Drug'] = nodes['Drug'].apply(lambda x: re.sub(r'\( *nasal *\)', '', x))

#show if there are () in the name of the drug
check = nodes[nodes['Drug'].str.contains('\(')]
check

	DDInterID	Drug	Drug_original
9	DDInter427	Cocaine (topical)	Cocaine (topical)
12	DDInter3	Abametapir (topical)	Abametapir (topical)
55	DDInter1333	Olopatadine (nasal)	Olopatadine (nasal)
60	DDInter764	Flurbiprofen (ophthalmic)	Flurbiprofen (ophthalmic)
100	DDInter154	Azelastine (nasal)	Azelastine (nasal)
...	...	...	...
3332	DDInter1225	Minocycline (topical)	Minocycline (topical)
3343	DDInter423	Coal tar (topical)	Coal tar (topical)
3344	DDInter861	Hexachlorophene (topical)	Hexachlorophene (topical)
3548	DDInter710	Factor XIII (human)	Factor XIII (human)
3620	DDInter1183	Methyl salicylate (topical)	Methyl salicylate (topical)

181 rows × 3 columns

nodes.shape

(1939, 3)

Add the market data to the Nodes Tables and then we drop the columns we do not need.

# rename Molecule_Kombi column to Drug
IQVIA.rename(columns={'Row Labels': 'Drug'}, inplace=True)

#lowercase all Drug names in IQVIA Table except for the first letter
IQVIA['Drug'] = IQVIA['Drug'].str.title()

# merge the two tables based , left join on nodes table
merged_df = pd.merge(nodes, IQVIA, on='Drug', how='left') 

# print the merged table
merged_df

	DDInterID	Drug	Drug_original	Units MAT Feb 2020	Units MAT Feb 2021	Units MAT Feb 2022	Units MAT Feb 2023	% Growth	Sum incl combi	Ranking Mono	Ranking Combi
0	DDInter513	Dexamethasone	Dexamethasone	359431.0	361420.0	380089.0	401082.0	0.055232	1162374.0	75.0	44.0
1	DDInter997	Ivacaftor	Ivacaftor	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
2	DDInter1019	Lamivudine	Lamivudine	8505.0	8570.0	6428.0	5231.0	-0.186217	57057.0	811.0	463.0
3	DDInter195	Betamethasone	Betamethasone	293676.0	259908.0	278059.0	294065.0	0.057563	877133.0	111.0	70.0
4	DDInter1961	Zidovudine	Zidovudine	708.0	670.0	459.0	413.0	-0.100218	1229.0	1086.0	1044.0
...	...	...	...	...	...	...	...	...	...	...	...
1934	DDInter371	Chondroitin sulfate	Chondroitin sulfate	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
1935	DDInter446	Cranberry	Cranberry	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
1936	DDInter829	Glucosamine	Glucosamine	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
1937	DDInter1390	Papaya	Papaya	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
1938	DDInter1551	Pyrophosphoric acid	Pyrophosphoric acid	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN

1939 rows × 11 columns

# enter zero for NaN value
merged_df = merged_df.fillna(0)

merged_df.columns

Index(['DDInterID', 'Drug', 'Drug_original', 'Units MAT Feb 2020',
       'Units  MAT Feb 2021', 'Units MAT Feb 2022', 'Units MAT Feb 2023',
       ' % Growth', 'Sum incl combi', 'Ranking Mono ', 'Ranking Combi'],
      dtype='object')

# drop the columns Units MAT Feb 2020,Units MAT Feb 2021,Units MAT Feb 2022
merged_df = merged_df.drop(['Units MAT Feb 2020','Units  MAT Feb 2021','Units MAT Feb 2022'], axis=1)

merged_df.head()

	DDInterID	Drug	Drug_original	Units MAT Feb 2023	% Growth	Sum incl combi	Ranking Mono	Ranking Combi
0	DDInter513	Dexamethasone	Dexamethasone	401082.0	0.055232	1162374.0	75.0	44.0
1	DDInter997	Ivacaftor	Ivacaftor	0.0	0.000000	0.0	0.0	0.0
2	DDInter1019	Lamivudine	Lamivudine	5231.0	-0.186217	57057.0	811.0	463.0
3	DDInter195	Betamethasone	Betamethasone	294065.0	0.057563	877133.0	111.0	70.0
4	DDInter1961	Zidovudine	Zidovudine	413.0	-0.100218	1229.0	1086.0	1044.0

nodes = merged_df

3.2. ADD Molecule Characteristics from CHEMBL

Adding attributes to the nodes table.

The columns describe relevant features of each pipe. The complete list is:

• oral - A binary flag indicating whether the molecule can be administered orally (i.e., by mouth)
• ophthalmic - A binary flag indicating whether the molecule can be administered ophthalmic( i.e. by eye)
• parenteral - A binary flag indicating whether the molecule can be administered by injection or other non-oral means
• topical - A binary flag indicating whether the molecule can be applied topically, i.e., to the skin
• chirality - A binary flag indicating whether the molecule is chiral, 0.0 Racemic Mixture, 1.0 Single Stereoisomer, 2.0 Archiral Moelcule
• black box warning - A binary flag indicating whether the drug has a black box warning, which is the strongest warning that the US Food and Drug Administration (FDA) can issue about a drug
• availability type - A classification of the type of availability of the drug, such as "1" (available for prescription), "2" (available over the counter), or "3" (withdrawn from market), - "-1" Undefined
• first in class - A binary flag indicating whether the molecule is the first in its class, i.e., the first drug to target a particular mechanism of action.

# Transform the Drug administrative name to a column 
merged_df['topical'] = merged_df['Drug_original'].apply(lambda x: 'topical' in x)
merged_df['nasal'] = merged_df['Drug_original'].apply(lambda x: 'nasal' in x)
merged_df['ophthalmic'] = merged_df['Drug_original'].apply(lambda x: 'ophthalmic' in x)

merged_df

	DDInterID	Drug	Drug_original	Units MAT Feb 2023	% Growth	Sum incl combi	Ranking Mono	Ranking Combi	topical	nasal	ophthalmic
0	DDInter513	Dexamethasone	Dexamethasone	401082.0	0.055232	1162374.0	75.0	44.0	False	False	False
1	DDInter997	Ivacaftor	Ivacaftor	0.0	0.000000	0.0	0.0	0.0	False	False	False
2	DDInter1019	Lamivudine	Lamivudine	5231.0	-0.186217	57057.0	811.0	463.0	False	False	False
3	DDInter195	Betamethasone	Betamethasone	294065.0	0.057563	877133.0	111.0	70.0	False	False	False
4	DDInter1961	Zidovudine	Zidovudine	413.0	-0.100218	1229.0	1086.0	1044.0	False	False	False
...	...	...	...	...	...	...	...	...	...	...	...
1934	DDInter371	Chondroitin sulfate	Chondroitin sulfate	0.0	0.000000	0.0	0.0	0.0	False	False	False
1935	DDInter446	Cranberry	Cranberry	0.0	0.000000	0.0	0.0	0.0	False	False	False
1936	DDInter829	Glucosamine	Glucosamine	0.0	0.000000	0.0	0.0	0.0	False	False	False
1937	DDInter1390	Papaya	Papaya	0.0	0.000000	0.0	0.0	0.0	False	False	False
1938	DDInter1551	Pyrophosphoric acid	Pyrophosphoric acid	0.0	0.000000	0.0	0.0	0.0	False	False	False

1939 rows × 11 columns

Add drug attributes to the nodes table

molecule = new_client.molecule
molecule_types = []

start = time.time()
processed = 0

for drug in nodes['Drug']:
    try:
        # compute duration to execute the loop
        # get current system time
        #print("processing: ", drug)
        #drug_start = time.time()
        
        # Remove content within brackets
        clean_drug = re.sub(r'\s?\(.*?\)', '', drug)
        
        mols = molecule.filter(molecule_synonyms__molecule_synonym__iexact=clean_drug).only(['oral'])
        molecule_types.append(mols[0]['oral'] if mols[0]['oral'] else '')
        #end = time.time()
        # print the time taken to execute the loop
        #processed += 1
        #print(f'Processed {processed} drugs, average processing time {(end - start)/processed}, last processing time {end - drug_start}')
    except:
        molecule_types.append('')

nodes['oral'] = molecule_types

molecule = new_client.molecule
molecule_types = []

start = time.time()
processed = 0

for drug in nodes['Drug']:
    try:
        # compute duration to execute the loop
        # get current system time
        #print("processing: ", drug)
        #drug_start = time.time()
        
        # Remove content within brackets
        clean_drug = re.sub(r'\s?\(.*?\)', '', drug)
        
        mols = molecule.filter(molecule_synonyms__molecule_synonym__iexact=clean_drug).only(['parenteral'])
        molecule_types.append(mols[0]['parenteral'] if mols[0]['parenteral'] else '')
        #end = time.time()
        # print the time taken to execute the loop
        #processed += 1
        #print(f'Processed {processed} drugs, average processing time {(end - start)/processed}, last processing time {end - drug_start}')
    except:
        molecule_types.append('')

nodes['parenteral'] = molecule_types

molecule = new_client.molecule
molecule_types = []

start = time.time()
processed = 0

for drug in nodes['Drug']:
    try:
        #print("processing: ", drug)
        drug_start = time.time()

        mols = molecule.filter(molecule_synonyms__molecule_synonym__iexact=drug).only(['topical'])
        molecule_types.append(mols[0]['topical'] if mols[0]['topical'] else '')
        end = time.time()
        # print the time taken to execute the loop
        #processed += 1
        #print(f'Processed {processed} drugs, average processing time {(end - start)/processed}, last processing time {end - drug_start}')
    except:
        molecule_types.append('')

nodes['topical2'] = molecule_types

#show columns that contains Drug name Dexamethasone
nodes[nodes['Drug'].str.contains('Dexamethasone')] 

	DDInterID	Drug	Drug_original	Units MAT Feb 2023	% Growth	Sum incl combi	Ranking Mono	Ranking Combi	topical	nasal	ophthalmic	oral	parenteral	topical2
0	DDInter513	Dexamethasone	Dexamethasone	401082.0	0.055232	1162374.0	75.0	44.0	False	False	False	True	True	True
618	DDInter515	Dexamethasone (ophthalmic)	Dexamethasone (ophthalmic)	0.0	0.000000	0.0	0.0	0.0	False	False	True	True	True
619	DDInter516	Dexamethasone (topical)	Dexamethasone (topical)	0.0	0.000000	0.0	0.0	0.0	True	False	False	True	True
689	DDInter514	Dexamethasone (nasal)	Dexamethasone (nasal)	0.0	0.000000	0.0	0.0	0.0	False	True	False	True	True

# for empty values in oral and parenteral columns, fill with False
nodes['oral'] = nodes['oral'].fillna(False)
nodes['parenteral'] = nodes['parenteral'].fillna(False)

# for topical, nasal and ophthalmic columns equal true fill oral and parenteral with False
nodes.loc[nodes['topical'] == True, 'oral'] = False
nodes.loc[nodes['topical'] == True, 'parenteral'] = False

nodes.loc[nodes['nasal'] == True, 'oral'] = False
nodes.loc[nodes['nasal'] == True, 'parenteral'] = False

nodes.loc[nodes['ophthalmic'] == True, 'oral'] = False
nodes.loc[nodes['ophthalmic'] == True, 'parenteral'] = False

# Replace topical2 ttrue values in topical column, since some data are missing in the drug inter database
import pandas as pd

# Assuming 'nodes' is your DataFrame
def update_topical(row):
    if row['topical2'] is True:
        return True
    else:
        return row['topical']

nodes['topical'] = nodes.apply(update_topical, axis=1)

# Drop the 'topical2' column as it is no longer needed
nodes = nodes.drop(columns=['topical2'])

#show columns that contains Drug name Dexamethasone
nodes[nodes['Drug'].str.contains('Olopatadine')] 

	DDInterID	Drug	Drug_original	Units MAT Feb 2023	% Growth	Sum incl combi	Ranking Mono	Ranking Combi	topical	nasal	ophthalmic	oral	parenteral
55	DDInter1333	Olopatadine (nasal)	Olopatadine (nasal)	0.0	0.000000	0.0	0.0	0.0	False	True	False	False	False
1149	DDInter1332	Olopatadine	Olopatadine	57545.0	0.051665	57545.0	354.0	458.0	True	False	False

molecule = new_client.molecule
ind_classes = []

start = time.time()
processed = 0

def clean_indications(indications):
    cleaned = set()
    for ind in indications.split(';'):
        ind = ind.strip()
        if ind not in cleaned:
            cleaned.add(ind)
    return '; '.join(cleaned)

for drug in nodes['Drug']:
    try:
        # print("processing: ", drug)
        # drug_start = time.time()
        
        # Remove content within brackets
        clean_drug = re.sub(r'\s?\(.*?\)', '', drug)
        
        mols = molecule.filter(molecule_synonyms__molecule_synonym__iexact=clean_drug).only('indication_class')
        ind_class = [mol['indication_class'] for mol in mols if mol['indication_class']]
        cleaned_ind_class = clean_indications('; '.join(ind_class))
        ind_classes.append(cleaned_ind_class)
        #end = time.time()
        # print the time taken to execute the loop
        #processed += 1
        #print(f'Processed {processed} drugs, average processing time {(end - start)/processed}, last processing time {end - drug_start}')
    except:
        ind_classes.append('')

nodes['indication_class'] = ind_classes

molecule = new_client.molecule
molecule_types = []

start = time.time()
processed = 0

for drug in nodes['Drug']:
    try:
        # compute duration to execute the loop
        # get current system time
        #print("processing: ", drug)
        #drug_start = time.time()
        
        # Remove content within brackets
        clean_drug = re.sub(r'\s?\(.*?\)', '', drug)
        
        mols = molecule.filter(molecule_synonyms__molecule_synonym__iexact=clean_drug).only(['chirality'])
        molecule_types.append(mols[0]['chirality'] if mols[0]['chirality'] else '')
        #end = time.time()
        # print the time taken to execute the loop
        #processed += 1
        #print(f'Processed {processed} drugs, average processing time {(end - start)/processed}, last processing time {end - drug_start}')
    except:
        molecule_types.append('')

nodes['chirality'] = molecule_types

molecule = new_client.molecule
molecule_types = []

start = time.time()
processed = 0

for drug in nodes['Drug']:
    try:
        # compute duration to execute the loop
        # get current system time
        #print("processing: ", drug)
        #drug_start = time.time()
        
        # Remove content within brackets
        clean_drug = re.sub(r'\s?\(.*?\)', '', drug)
        
        mols = molecule.filter(molecule_synonyms__molecule_synonym__iexact=clean_drug).only(['black_box_warning'])
        molecule_types.append(mols[0]['black_box_warning'] if mols[0]['black_box_warning'] else '')
        #end = time.time()
        # print the time taken to execute the loop
        #processed += 1
        #print(f'Processed {processed} drugs, average processing time {(end - start)/processed}, last processing time {end - drug_start}')
    except:
        molecule_types.append('')

nodes['black_box_warning'] = molecule_types

molecule = new_client.molecule
molecule_types = []

start = time.time()
processed = 0

for drug in nodes['Drug']:
    try:
        # compute duration to execute the loop
        # get current system time
        #print("processing: ", drug)
        #drug_start = time.time()
        
        # Remove content within brackets
        clean_drug = re.sub(r'\s?\(.*?\)', '', drug)
        
        mols = molecule.filter(molecule_synonyms__molecule_synonym__iexact=clean_drug).only(['availability_type'])
        molecule_types.append(mols[0]['availability_type'] if mols[0]['availability_type'] else '')
        #end = time.time()
        # print the time taken to execute the loop
        #processed += 1
        #print(f'Processed {processed} drugs, average processing time {(end - start)/processed}, last processing time {end - drug_start}')
    except:
        molecule_types.append('')

nodes['availability_type'] = molecule_types

molecule = new_client.molecule
molecule_types = []

start = time.time()
processed = 0

for drug in nodes['Drug']:
    try:
        # compute duration to execute the loop
        # get current system time
        #print("processing: ", drug)
        #drug_start = time.time()
        
        # Remove content within brackets
        clean_drug = re.sub(r'\s?\(.*?\)', '', drug)
        
        mols = molecule.filter(molecule_synonyms__molecule_synonym__iexact=clean_drug).only(['first_in_class'])
        molecule_types.append(mols[0]['first_in_class'] if mols[0]['first_in_class'] else '')
        #end = time.time()
        # print the time taken to execute the loop
        #processed += 1
        #print(f'Processed {processed} drugs, average processing time {(end - start)/processed}, last processing time {end - drug_start}')
    except:
        molecule_types.append('')

nodes['first_in_class'] = molecule_types

nodes.head()

	DDInterID	Drug	Drug_original	Units MAT Feb 2023	% Growth	Sum incl combi	Ranking Mono	Ranking Combi	topical	nasal	ophthalmic	oral	parenteral	indication_class	chirality	black_box_warning	availability_type	first_in_class
0	DDInter513	Dexamethasone	Dexamethasone	401082.0	0.055232	1162374.0	75.0	44.0	True	False	False	True	True	Glucocorticoid	1		1
1	DDInter997	Ivacaftor	Ivacaftor	0.0	0.000000	0.0	0.0	0.0	False	False	False	True			2		1	1
2	DDInter1019	Lamivudine	Lamivudine	5231.0	-0.186217	57057.0	811.0	463.0	False	False	False	True		Antiviral	1	1	1
3	DDInter195	Betamethasone	Betamethasone	294065.0	0.057563	877133.0	111.0	70.0	True	False	False	True		Glucocorticoid	1
4	DDInter1961	Zidovudine	Zidovudine	413.0	-0.100218	1229.0	1086.0	1044.0	False	False	False	True	True	Antiviral	1	1	1

3.3. ATC classification

In the Anatomical Therapeutic Chemical (ATC) classification system, the active substances are divided into different groups according to the organ or system on which they act and their therapeutic, pharmacological and chemical properties. Drugs are classified in groups at five different levels.

The 2nd, 3rd and 4th levels are often used to identify pharmacological subgroups when that is considered more appropriate than therapeutic or chemical subgroups.

3.3.1. ATC code classification

Adding all ATC codes for each drug.

#Adding ATC classes to the nodes
molecule = new_client.molecule
atc_classes = []

for drug in nodes['Drug']:
    try:
        # compute duration to execute the loop
        # get current system time
        #print("processing: ", drug)
        #drug_start = time.time()

        # Remove content within brackets
        clean_drug = re.sub(r'\s?\(.*?\)', '', drug)

        mols = molecule.filter(molecule_synonyms__molecule_synonym__iexact=clean_drug).only('atc_classifications')
        atc_classes.append('; '.join(mols[0]['atc_classifications']) if mols[0]['atc_classifications'] else '')
        #end = time.time()
        # print the time taken to execute the loop
        #processed += 1
        #print(f'Processed {processed} drugs, average processing time {(end - start)/processed}, last processing time {end - drug_start}')
    except:
        atc_classes.append('')

nodes['ATC_Classifications'] = atc_classes

# assume nodes is a pandas DataFrame containing 'ATC_Classifications' and 'Drug' columns
empty_cols = nodes.isnull().any()

if empty_cols['ATC_Classifications']:
    pd.options.display.max_rows = None
    print(nodes[nodes['ATC_Classifications'].isnull()][['Drug', 'ATC_Classifications']])

#Check for empty rows
empty_rows = nodes[nodes['ATC_Classifications'] == '']
if not empty_rows.empty:
    print(empty_rows[['Drug', 'ATC_Classifications']])

                               Drug ATC_Classifications
9                 Cocaine (topical)                    
41           Botulinum toxin type A                    
68                     Butabarbital                    
69                       Butalbital                    
111   Carbamoylcholine (ophthalmic)                    
...                             ...                 ...
1932    Methyl salicylate (topical)                    
1933           Citrus bioflavonoids                    
1935                      Cranberry                    
1937                         Papaya                    
1938            Pyrophosphoric acid                    

[312 rows x 2 columns]

nodes

	DDInterID	Drug	Drug_original	Units MAT Feb 2023	% Growth	Sum incl combi	Ranking Mono	Ranking Combi	topical	nasal	ophthalmic	oral	parenteral	indication_class	chirality	black_box_warning	availability_type	first_in_class	ATC_Classifications
0	DDInter513	Dexamethasone	Dexamethasone	401082.0	0.055232	1162374.0	75.0	44.0	True	False	False	True	True	Glucocorticoid	1		1		R01AD53; A01AC02; R01AD03; S02BA06; S03BA01; D...
1	DDInter997	Ivacaftor	Ivacaftor	0.0	0.000000	0.0	0.0	0.0	False	False	False	True			2		1	1	R07AX02
2	DDInter1019	Lamivudine	Lamivudine	5231.0	-0.186217	57057.0	811.0	463.0	False	False	False	True		Antiviral	1	1	1		J05AF05
3	DDInter195	Betamethasone	Betamethasone	294065.0	0.057563	877133.0	111.0	70.0	True	False	False	True		Glucocorticoid	1				D07AC01; D07XC01; H02AB01; S02BA07; S01CB04; S...
4	DDInter1961	Zidovudine	Zidovudine	413.0	-0.100218	1229.0	1086.0	1044.0	False	False	False	True	True	Antiviral	1	1	1		J05AF01
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
1934	DDInter371	Chondroitin sulfate	Chondroitin sulfate	0.0	0.000000	0.0	0.0	0.0	False	False	False				1		-1		M01AX25
1935	DDInter446	Cranberry	Cranberry	0.0	0.000000	0.0	0.0	0.0	False	False	False
1936	DDInter829	Glucosamine	Glucosamine	0.0	0.000000	0.0	0.0	0.0	False	False	False				1		-1		M01AX05
1937	DDInter1390	Papaya	Papaya	0.0	0.000000	0.0	0.0	0.0	False	False	False
1938	DDInter1551	Pyrophosphoric acid	Pyrophosphoric acid	0.0	0.000000	0.0	0.0	0.0	False	False	False

1939 rows × 19 columns

3.3.2 Adding ATC1 code to the table

# Create a dictionary from the matching table
atc_dict = dict(zip(ATC['atc_code'], ATC['atc_name']))

# Define a function to lookup the ATC names
def lookup_names(atc_codes):
    atc_codes = atc_codes.split('; ')
    atc_codes = [code[:1] for code in atc_codes]
    atc_count = {}
    for code in atc_codes:
        if code in atc_dict:
            atc_name = atc_dict[code]
            if atc_name in atc_count:
                atc_count[atc_name] += 1
            else:
                atc_count[atc_name] = 1
    return atc_count

# Apply the function to create the new column
nodes['ATC1_names'] = nodes['ATC_Classifications'].apply(lookup_names)

nodes.head(2)

	DDInterID	Drug	Drug_original	Units MAT Feb 2023	% Growth	Sum incl combi	Ranking Mono	Ranking Combi	topical	nasal	ophthalmic	oral	parenteral	indication_class	chirality	black_box_warning	availability_type	first_in_class	ATC_Classifications	ATC1_names
0	DDInter513	Dexamethasone	Dexamethasone	401082.0	0.055232	1162374.0	75.0	44.0	True	False	False	True	True	Glucocorticoid	1		1		R01AD53; A01AC02; R01AD03; S02BA06; S03BA01; D...	{'RESPIRATORY SYSTEM': 2, 'ALIMENTARY TRACT AN...
1	DDInter997	Ivacaftor	Ivacaftor	0.0	0.000000	0.0	0.0	0.0	False	False	False	True			2		1	1	R07AX02	{'RESPIRATORY SYSTEM': 1}

#Without the warning
# Step 1: Create a set of all unique ATC1 names
unique_atc_names = set()
for row in nodes.itertuples():
    unique_atc_names.update(row.ATC1_names.keys())

# Step 2: Create a new DataFrame with all the new columns
new_cols = {}
for atc_name in unique_atc_names:
    col_name = atc_name.replace(' ', '_')
    new_cols[col_name] = nodes.apply(lambda row: row.ATC1_names.get(atc_name, 0), axis=1)

# Combine the new columns with the original DataFrame
nodes = pd.concat([nodes, pd.DataFrame(new_cols)], axis=1)

# Drop the original ATC1_names column
#nodes = nodes.drop(columns=['ATC1_names'])

nodes.columns

Index(['DDInterID', 'Drug', 'Drug_original', 'Units MAT Feb 2023', ' % Growth',
       'Sum incl combi', 'Ranking Mono ', 'Ranking Combi', 'topical', 'nasal',
       'ophthalmic', 'oral', 'parenteral', 'indication_class', 'chirality',
       'black_box_warning', 'availability_type', 'first_in_class',
       'ATC_Classifications', 'ATC1_names', 'SENSORY_ORGANS',
       'DERMATOLOGICALS',
       'SYSTEMIC_HORMONAL_PREPARATIONS,_EXCL._SEX_HORMONES_AND_INSULINS',
       'ANTIPARASITIC_PRODUCTS,_INSECTICIDES_AND_REPELLENTS',
       'CARDIOVASCULAR_SYSTEM', 'ANTINEOPLASTIC_AND_IMMUNOMODULATING_AGENTS',
       'GENITO_URINARY_SYSTEM_AND_SEX_HORMONES', 'MUSCULO-SKELETAL_SYSTEM',
       'RESPIRATORY_SYSTEM', 'NERVOUS_SYSTEM',
       'ALIMENTARY_TRACT_AND_METABOLISM', 'ANTIINFECTIVES_FOR_SYSTEMIC_USE',
       'BLOOD_AND_BLOOD_FORMING_ORGANS', 'VARIOUS'],
      dtype='object')

3.3.3 Adding ATC2 code to the table

Since we add 88 more columns with ATC2 code, we will work with the ATC1 and ATC2 for the moment

def lookup_atc2(atc_codes):
    atc_codes = atc_codes.split('; ')
    atc2_codes = []
    for code in atc_codes:
        if code[:3] in atc_dict:
            atc_name = atc_dict[code[:3]]
            if atc_name.startswith('V') or atc_name.startswith('M'):
                continue
            atc2_codes.append(f'{code[:3]} - {atc_name}')
    return '; '.join(atc2_codes)

# Add the new ATC2 column
nodes['ATC2_names'] = nodes['ATC_Classifications'].apply(lookup_atc2)

def lookup_names_with_prefix(atc_codes):
    atc_codes = atc_codes.split('; ')
    atc_dict_with_prefix = {}
    for code in atc_codes:
        if code[:3] in atc_dict:
            atc_dict_with_prefix[code[:3] + ' - ' + atc_dict[code[:3]]] = 1
    return atc_dict_with_prefix

# Add the new ATC2 column
nodes['ATC2_names'] = nodes['ATC_Classifications'].apply(lookup_names_with_prefix)

pd.set_option('display.max_columns', None)
nodes.head(3)

	DDInterID	Drug	Drug_original	Units MAT Feb 2023	% Growth	Sum incl combi	Ranking Mono	Ranking Combi	topical	nasal	ophthalmic	oral	parenteral	indication_class	chirality	black_box_warning	availability_type	first_in_class	ATC_Classifications	ATC1_names	SENSORY_ORGANS	DERMATOLOGICALS	SYSTEMIC_HORMONAL_PREPARATIONS,_EXCL._SEX_HORMONES_AND_INSULINS	ANTIPARASITIC_PRODUCTS,_INSECTICIDES_AND_REPELLENTS	CARDIOVASCULAR_SYSTEM	ANTINEOPLASTIC_AND_IMMUNOMODULATING_AGENTS	GENITO_URINARY_SYSTEM_AND_SEX_HORMONES	MUSCULO-SKELETAL_SYSTEM	RESPIRATORY_SYSTEM	NERVOUS_SYSTEM	ALIMENTARY_TRACT_AND_METABOLISM	ANTIINFECTIVES_FOR_SYSTEMIC_USE	BLOOD_AND_BLOOD_FORMING_ORGANS	VARIOUS	ATC2_names
0	DDInter513	Dexamethasone	Dexamethasone	401082.0	0.055232	1162374.0	75.0	44.0	True	False	False	True	True	Glucocorticoid	1		1		R01AD53; A01AC02; R01AD03; S02BA06; S03BA01; D...	{'RESPIRATORY SYSTEM': 2, 'ALIMENTARY TRACT AN...	4	3	1	0	1	0	0	0	2	0	1	0	0	0	{'R01 - NASAL PREPARATIONS': 1, 'A01 - STOMATO...
1	DDInter997	Ivacaftor	Ivacaftor	0.0	0.000000	0.0	0.0	0.0	False	False	False	True			2		1	1	R07AX02	{'RESPIRATORY SYSTEM': 1}	0	0	0	0	0	0	0	0	1	0	0	0	0	0	{'R07 - OTHER RESPIRATORY SYSTEM PRODUCTS': 1}
2	DDInter1019	Lamivudine	Lamivudine	5231.0	-0.186217	57057.0	811.0	463.0	False	False	False	True		Antiviral	1	1	1		J05AF05	{'ANTIINFECTIVES FOR SYSTEMIC USE': 1}	0	0	0	0	0	0	0	0	0	0	0	1	0	0	{'J05 - ANTIVIRALS FOR SYSTEMIC USE': 1}

# Step 1: Create a dictionary of all unique ATC2 names with prefixes
unique_atc_names_with_prefix = {}
for row in nodes.itertuples():
    unique_atc_names_with_prefix.update(row.ATC2_names)

# Step 2: Create new columns for each unique ATC2 name with prefix
for atc_name in unique_atc_names_with_prefix:
    nodes[atc_name.replace(' ', '_')] = 0

# Step 3: Fill the new columns with the corresponding values from the ATC_Classifications column
for row in nodes.itertuples():
    for atc_name in row.ATC2_names:
        nodes.at[row.Index, atc_name.replace(' ', '_')] = 1

# Drop the original ATC_Classifications column
#nodes_test = nodes.drop(columns=['ATC2_names'])

nodes.columns

Index(['DDInterID', 'Drug', 'Drug_original', 'Units MAT Feb 2023', ' % Growth',
       'Sum incl combi', 'Ranking Mono ', 'Ranking Combi', 'topical', 'nasal',
       ...
       'V09_-_DIAGNOSTIC_RADIOPHARMACEUTICALS', 'H03_-_THYROID_THERAPY',
       'B03_-_ANTIANEMIC_PREPARATIONS', 'D09_-_MEDICATED_DRESSINGS',
       'D03_-_PREPARATIONS_FOR_TREATMENT_OF_WOUNDS_AND_ULCERS',
       'D02_-_EMOLLIENTS_AND_PROTECTIVES', 'J07_-_VACCINES',
       'A09_-_DIGESTIVES,_INCL._ENZYMES', 'V06_-_GENERAL_NUTRIENTS',
       'H04_-_PANCREATIC_HORMONES'],
      dtype='object', length=123)

# Replace commas and semicolons with pipes
nodes = nodes.replace([';'], ',', regex=True)

#export to csv with '|' as delimiter
nodes.to_csv('/Users/stephaniegessler/IE Students/Term 3 - Data Dynamos - General/Social Network Analysis/drugs/nodes.csv', index=False, sep=';')

4. Graph Statistics

4.1 Graph Summary

import pandas as pd
# import data from csv nodes file
nodes = pd.read_csv('/Users/stephaniegessler/IE Students/Term 3 - Data Dynamos - General/Social Network Analysis/drugs/nodes.csv', sep=';')
edges = pd.read_csv('/Users/stephaniegessler/IE Students/Term 3 - Data Dynamos - General/Social Network Analysis/drugs/edges.csv', sep=';')

# Replace all NANs with zero in table nodes 
nodes = nodes.fillna(0)

edges

	DDInterID_A	Drug_A	DDInterID_B	Drug_B	Level	Weight
0	DDInter513	Dexamethasone	DDInter582	Dolutegravir	Minor	3
1	DDInter997	Ivacaftor	DDInter582	Dolutegravir	Minor	3
2	DDInter1019	Lamivudine	DDInter1123	Mannitol	Moderate	2
3	DDInter195	Betamethasone	DDInter1961	Zidovudine	Moderate	2
4	DDInter1961	Zidovudine	DDInter246	Budesonide	Moderate	2
...	...	...	...	...	...	...
160230	DDInter856	Heparin	DDInter805	gamma-Aminobutyric acid	Unknown	2
160231	DDInter1683	Sodium bicarbonate	DDInter805	gamma-Aminobutyric acid	Unknown	2
160232	DDInter908	Iloprost	DDInter62	Ambrisentan	Unknown	2
160233	DDInter1951	Warfarin	DDInter1942	Vitamin D	Unknown	2
160234	DDInter1683	Sodium bicarbonate	DDInter1942	Vitamin D	Unknown	2

160235 rows × 6 columns

Checking the distribution of the Weights, where you can observe that moderate has the highest interaction weights.

import seaborn as sns
# Create a countplot of the Level column
sns.countplot(x='Level', data=edges)

<AxesSubplot: xlabel='Level', ylabel='count'>

#create graph with networkx weighted 
G = nx.from_pandas_edgelist(edges, 'DDInterID_A', 'DDInterID_B', 'Weight', create_using=nx.Graph())

#add all attributes from nodes to graph
#nx.set_node_attributes(G, nodes.set_index('Drug_original').to_dict('index'))
node_dict = {d.pop('Drug_original'): d for d in nodes.to_dict('records')}
nx.set_node_attributes(G, node_dict)

Checking the Graph summary, we have 1939 nodes and 160235 Edges,

print("Number of nodes:", G.number_of_nodes())
print("Number of edges:", G.number_of_edges())

Number of nodes: 1939
Number of edges: 160235

We have 3 Subgraphs.

# get all the subgraphs
subgraphs = nx.connected_components(G)

# get the count of subgraphs
subgraph_count = len(list(subgraphs))

print("Subgraph count:", subgraph_count)

Subgraph count: 3

4.2 ATC Graphics and Distribution

ATC1 Distribution

Checking the distribution of the drugs, where nervous system has the highest distribution with 319 drug, then Antineoplastic second 298 drugs and third Alimentary Tract and Metabalism with 273 drugs.

import plotly.graph_objs as go

# Define the mapping between column names and labels
col_to_label = {'RESPIRATORY_SYSTEM': 'R',
                'CARDIOVASCULAR_SYSTEM': 'C',
                'ANTIPARASITIC_PRODUCTS,_INSECTICIDES_AND_REPELLENTS': 'P',
                'BLOOD_AND_BLOOD_FORMING_ORGANS': 'B',
                'MUSCULO-SKELETAL_SYSTEM': 'M',
                'ANTINEOPLASTIC_AND_IMMUNOMODULATING_AGENTS': 'L',
                'DERMATOLOGICALS': 'D',
                'GENITO_URINARY_SYSTEM_AND_SEX_HORMONES': 'G',
                'SYSTEMIC_HORMONAL_PREPARATIONS,_EXCL._SEX_HORMONES_AND_INSULINS': 'H',
                'SENSORY_ORGANS': 'S',
                'VARIOUS': 'V',
                'ALIMENTARY_TRACT_AND_METABOLISM': 'A',
                'ANTIINFECTIVES_FOR_SYSTEMIC_USE': 'J',
                'NERVOUS_SYSTEM': 'N'}

# Define the colors for each column
colors = ['blue', 'red', 'green', 'purple', 'orange', 'gray', 'pink', 'brown', 'teal', 'olive', 'navy', 'maroon', 'cyan', 'magenta']

# Sort the columns by their sum in descending order
nodes2 = nodes.iloc[:, 20:-89].assign(total=nodes.iloc[:, 20:-89].sum(axis=1)).sort_values('total', ascending=False).drop('total', axis=1)

# Create the bar chart
fig = go.Figure()
for i, col in enumerate(nodes2.columns):
    fig.add_trace(go.Bar(x=[nodes2[col].sum()], y=[col_to_label[col]], name=col, orientation='h', marker_color=colors[i], text=[nodes2[col].sum()], textposition='outside'))

# Set the title and axis labels
fig.update_layout(title='Sum of values for each system', xaxis_title='Sum', yaxis_title='System', yaxis_categoryorder='total ascending')

# Update the legend
#fig.update_layout(legend=dict(title='ATC Classifications', orientation='h', yanchor='top', y=1.05, xanchor='left', x=0.1))

# Set the title and axis labels
fig.update_layout(title='Sum of each Drug per ATC1', xaxis_title='Sum', yaxis_title='System',
                  legend=dict(yanchor="top", y=0.99, xanchor="right", x=1.8))

# Show the plot
fig.show()

Top 10 ATC2 Distribution

The highest Drug Distribution is under the Therapeutic area: Antineoplastic Agents, then Ophtalmologicals and then Antibacterials for Systemic use.

import plotly.graph_objs as go

# Define the colors for each column
colors = ['blue', 'red', 'green', 'purple', 'orange', 'gray', 'pink', 'brown', 'teal', 'olive', 'navy', 'maroon', 'cyan', 'magenta']

# Get the top 20 columns
top_columns = nodes.iloc[:, 35:].sum(axis=0).sort_values(ascending=False).head(10)

# Create the bar chart
fig = go.Figure()
for i, col in enumerate(top_columns.index):
    fig.add_trace(go.Bar(x=[top_columns[col]], y=[col], name=col, orientation='h', marker_color=colors[i%len(colors)], text=[top_columns[col]], textposition='outside'))

# Set the title and axis labels
fig.update_layout(title='Top 10 ATC2 Classifications', xaxis_title='Total Number of Molecules', yaxis_title='ATC2 Classidication')

# Show the plot
fig.show()

# Get the subgraphs
subgraphs = nx.connected_components(G)

__Diameter__

# Check if the graph is connected
if not nx.is_connected(G):
    # If not, get the connected components and calculate the diameter of each component
    subgraphs = list(nx.connected_components(G))
    for i, sg in enumerate(subgraphs):
        try:
            diameter = nx.algorithms.distance_measures.diameter(G.subgraph(sg))
            print(f"Diameter of subgraph {i+1}: {diameter}")
        except nx.exception.NetworkXError as e:
            print(f"Subgraph {i+1} is not connected, so it does not have a diameter")
else:
    # If connected, calculate the diameter of the entire graph
    try:
        diameter = nx.algorithms.distance_measures.diameter(G)
        print(f"The diameter of the graph is {diameter}")
    except nx.exception.NetworkXError as e:
        print("The graph is not connected, so it does not have a diameter")

Diameter of subgraph 1: 7
Diameter of subgraph 2: 2
Diameter of subgraph 3: 1

5.3 Pagerank

The diameter of a graph is the maximum distance between any two pairs of nodes in the graph. In this case, a diameter of 7 means that the longest shortest path between any two nodes in your graph is 7. The other are very short and not densely connected graph with a diameter of 2 and 1 respectivley.

This information can be useful for understanding the overall structure and connectedness of your graph. A small diameter (shortest path between any two nodes) indicates that the graph is well connected, whereas a large diameter may indicate that there are some isolated nodes or subgraphs within the graph.

# Check if the graph is connected
if not nx.is_connected(G):
    # If not, get the connected components and calculate the average shortest path, max and min degree of each component
    subgraphs = list(nx.connected_components(G))
    for i, sg in enumerate(subgraphs):
        # Calculate average shortest path length
        shortest_path_avg = nx.average_shortest_path_length(G.subgraph(sg))
        print(f"Shortest average path length of subgraph {i+1}: {shortest_path_avg}")
        
        # Calculate max and min degree
        max_degree = max(dict(G.subgraph(sg).degree()).values())
        min_degree = min(dict(G.subgraph(sg).degree()).values())
        print(f"Maximum degree of subgraph {i+1}: {max_degree}")
        print(f"Minimum degree of subgraph {i+1}: {min_degree}")
else:
    # If connected, calculate the shortest average path, max and min degree of the entire graph
    shortest_path_avg = nx.average_shortest_path_length(G)
    print(f"The shortest average path length is {shortest_path_avg}")
    
    max_degree = max(dict(G.degree()).values())
    min_degree = min(dict(G.degree()).values())
    print(f"Maximum degree of the graph: {max_degree}")
    print(f"Minimum degree of the graph: {min_degree}")

Shortest average path length of subgraph 1: 2.180150876492809
Maximum degree of subgraph 1: 910
Minimum degree of subgraph 1: 1
Shortest average path length of subgraph 2: 1.3333333333333333
Maximum degree of subgraph 2: 2
Minimum degree of subgraph 2: 1
Shortest average path length of subgraph 3: 1.0
Maximum degree of subgraph 3: 1
Minimum degree of subgraph 3: 1

The first set of results pertains to subgraph 1, which has an average shortest path length of 2.18. This suggests that, on average, it takes 2.18 steps to get from one node in subgraph 1 to another node in the same subgraph. The maximum degree of subgraph 1 is 910, which means that at least one node in subgraph 1 has 910 connections. The minimum degree of subgraph 1 is 1, indicating that there is at least one node in subgraph 1 with only one connection.

The second set of results relates to subgraph 2, which has an average shortest path length of 1.33. This indicates that it takes only 1.33 steps on average to get from one node to another node in subgraph 2. The maximum degree of subgraph 2 is 2, meaning that no node in subgraph 2 has more than two connections. The minimum degree of subgraph 2 is also 1, suggesting that there is at least one node in subgraph 2 with only one connection.

Finally, the third set of results pertains to subgraph 3, which has an average shortest path length of 1.0. This means that it takes only one step to get from any node in subgraph 3 to any other node in the same subgraph. The maximum degree of subgraph 3 is 1, indicating that every node in subgraph 3 has only one connection. The minimum degree of subgraph 3 is also 1, implying that every node in subgraph 3 has at least one connection.

# get the connected components
subgraphs = list(nx.connected_components(G))

# plot each subgraph vertically
fig, axs = plt.subplots(len(subgraphs), 1, figsize=(5, 3*len(subgraphs)))
for i, sg in enumerate(subgraphs):
    axs[i].set_title(f"Subgraph {i+1}")
    sg = G.subgraph(sg)
    pos = nx.spring_layout(sg)
    nx.draw(sg, pos, ax=axs[i], node_size=500)

plt.tight_layout()
plt.show()

# Map the subgraphs in the nodes DataFrame
subgraph_dict = {}
for i, subgraph in enumerate(nx.connected_components(G)):
    for node in subgraph:
        subgraph_dict[node] = i+1

nodes['subgraph'] = nodes['DDInterID'].map(subgraph_dict)

# show all columns
pd.set_option('display.max_columns', None)
nodes.head(2)

	DDInterID	Drug	Drug_original	Units MAT Feb 2023	% Growth	Sum incl combi	Ranking Mono	Ranking Combi	topical	nasal	ophthalmic	oral	parenteral	indication_class	chirality	black_box_warning	availability_type	first_in_class	ATC_Classifications	ATC1_names	SENSORY_ORGANS	DERMATOLOGICALS	SYSTEMIC_HORMONAL_PREPARATIONS,_EXCL._SEX_HORMONES_AND_INSULINS	ANTIPARASITIC_PRODUCTS,_INSECTICIDES_AND_REPELLENTS	CARDIOVASCULAR_SYSTEM	ANTINEOPLASTIC_AND_IMMUNOMODULATING_AGENTS	GENITO_URINARY_SYSTEM_AND_SEX_HORMONES	MUSCULO-SKELETAL_SYSTEM	RESPIRATORY_SYSTEM	NERVOUS_SYSTEM	ALIMENTARY_TRACT_AND_METABOLISM	ANTIINFECTIVES_FOR_SYSTEMIC_USE	BLOOD_AND_BLOOD_FORMING_ORGANS	VARIOUS	ATC2_names	R01_-_NASAL_PREPARATIONS	A01_-_STOMATOLOGICAL_PREPARATIONS	S02_-_OTOLOGICALS	S03_-_OPHTHALMOLOGICAL_AND_OTOLOGICAL_PREPARATIONS	D07_-_CORTICOSTEROIDS,_DERMATOLOGICAL_PREPARATIONS	D10_-_ANTI-ACNE_PREPARATIONS	S01_-_OPHTHALMOLOGICALS	C05_-_VASOPROTECTIVES	H02_-_CORTICOSTEROIDS_FOR_SYSTEMIC_USE	R07_-_OTHER_RESPIRATORY_SYSTEM_PRODUCTS	J05_-_ANTIVIRALS_FOR_SYSTEMIC_USE	R03_-_DRUGS_FOR_OBSTRUCTIVE_AIRWAY_DISEASES	A07_-_ANTIDIARRHEALS,_INTESTINAL_ANTIINFLAMMATORY/ANTIINFECTIVE_AGENTS	R06_-_ANTIHISTAMINES_FOR_SYSTEMIC_USE	N06_-_PSYCHOANALEPTICS	D11_-_OTHER_DERMATOLOGICAL_PREPARATIONS	N02_-_ANALGESICS	R05_-_COUGH_AND_COLD_PREPARATIONS	N07_-_OTHER_NERVOUS_SYSTEM_DRUGS	P03_-_ECTOPARASITICIDES,_INCL._SCABICIDES,_INSECTICIDES_AND_REPELLENTS	R02_-_THROAT_PREPARATIONS	D04_-_ANTIPRURITICS,_INCL._ANTIHISTAMINES,_ANESTHETICS,_ETC.	N01_-_ANESTHETICS	C01_-_CARDIAC_THERAPY	C04_-_PERIPHERAL_VASODILATORS	L04_-_IMMUNOSUPPRESSANTS	B01_-_ANTITHROMBOTIC_AGENTS	L02_-_ENDOCRINE_THERAPY	L01_-_ANTINEOPLASTIC_AGENTS	A11_-_VITAMINS	M02_-_TOPICAL_PRODUCTS_FOR_JOINT_AND_MUSCULAR_PAIN	M01_-_ANTIINFLAMMATORY_AND_ANTIRHEUMATIC_PRODUCTS	G02_-_OTHER_GYNECOLOGICALS	A10_-_DRUGS_USED_IN_DIABETES	C07_-_BETA_BLOCKING_AGENTS	J02_-_ANTIMYCOTICS_FOR_SYSTEMIC_USE	D01_-_ANTIFUNGALS_FOR_DERMATOLOGICAL_USE	N05_-_PSYCHOLEPTICS	H01_-_PITUITARY_AND_HYPOTHALAMIC_HORMONES_AND_ANALOGUES	N03_-_ANTIEPILEPTICS	M03_-_MUSCLE_RELAXANTS	N04_-_ANTI-PARKINSON_DRUGS	G04_-_UROLOGICALS	A03_-_DRUGS_FOR_FUNCTIONAL_GASTROINTESTINAL_DISORDERS	C02_-_ANTIHYPERTENSIVES	A04_-_ANTIEMETICS_AND_ANTINAUSEANTS	D08_-_ANTISEPTICS_AND_DISINFECTANTS	V03_-_ALL_OTHER_THERAPEUTIC_PRODUCTS	A08_-_ANTIOBESITY_PREPARATIONS,_EXCL._DIET_PRODUCTS	A12_-_MINERAL_SUPPLEMENTS	B05_-_BLOOD_SUBSTITUTES_AND_PERFUSION_SOLUTIONS	B02_-_ANTIHEMORRHAGICS	L03_-_IMMUNOSTIMULANTS	C03_-_DIURETICS	C09_-_AGENTS_ACTING_ON_THE_RENIN-ANGIOTENSIN_SYSTEM	A02_-_DRUGS_FOR_ACID_RELATED_DISORDERS	J01_-_ANTIBACTERIALS_FOR_SYSTEMIC_USE	D06_-_ANTIBIOTICS_AND_CHEMOTHERAPEUTICS_FOR_DERMATOLOGICAL_USE	A06_-_DRUGS_FOR_CONSTIPATION	P01_-_ANTIPROTOZOALS	J04_-_ANTIMYCOBACTERIALS	V04_-_DIAGNOSTIC_AGENTS	G03_-_SEX_HORMONES_AND_MODULATORS_OF_THE_GENITAL_SYSTEM	M09_-_OTHER_DRUGS_FOR_DISORDERS_OF_THE_MUSCULO-SKELETAL_SYSTEM	G01_-_GYNECOLOGICAL_ANTIINFECTIVES_AND_ANTISEPTICS	C08_-_CALCIUM_CHANNEL_BLOCKERS	M05_-_DRUGS_FOR_TREATMENT_OF_BONE_DISEASES	V08_-_CONTRAST_MEDIA	C10_-_LIPID_MODIFYING_AGENTS	V10_-_THERAPEUTIC_RADIOPHARMACEUTICALS	D05_-_ANTIPSORIATICS	A16_-_OTHER_ALIMENTARY_TRACT_AND_METABOLISM_PRODUCTS	M04_-_ANTIGOUT_PREPARATIONS	B06_-_OTHER_HEMATOLOGICAL_AGENTS	A14_-_ANABOLIC_AGENTS_FOR_SYSTEMIC_USE	P02_-_ANTHELMINTICS	H05_-_CALCIUM_HOMEOSTASIS	A05_-_BILE_AND_LIVER_THERAPY	V09_-_DIAGNOSTIC_RADIOPHARMACEUTICALS	H03_-_THYROID_THERAPY	B03_-_ANTIANEMIC_PREPARATIONS	D09_-_MEDICATED_DRESSINGS	D03_-_PREPARATIONS_FOR_TREATMENT_OF_WOUNDS_AND_ULCERS	D02_-_EMOLLIENTS_AND_PROTECTIVES	J07_-_VACCINES	A09_-_DIGESTIVES,_INCL._ENZYMES	V06_-_GENERAL_NUTRIENTS	H04_-_PANCREATIC_HORMONES	subgraph
0	DDInter513	Dexamethasone	Dexamethasone	401082.0	0.055232	1162374.0	75.0	44.0	True	False	False	True	True	Glucocorticoid	1.0	0.0	1.0	0.0	R01AD53, A01AC02, R01AD03, S02BA06, S03BA01, D...	{'RESPIRATORY SYSTEM': 2, 'ALIMENTARY TRACT AN...	4	3	1	0	1	0	0	0	2	0	1	0	0	0	{'R01 - NASAL PREPARATIONS': 1, 'A01 - STOMATO...	1	1	1	1	1	1	1	1	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1
1	DDInter997	Ivacaftor	Ivacaftor	0.0	0.000000	0.0	0.0	0.0	False	False	False	True	0	0	2.0	0.0	1.0	1.0	R07AX02	{'RESPIRATORY SYSTEM': 1}	0	0	0	0	0	0	0	0	1	0	0	0	0	0	{'R07 - OTHER RESPIRATORY SYSTEM PRODUCTS': 1}	0	0	0	0	0	0	0	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1

#Replace 0 with false for oral and parenteral,topical
nodes['oral'] = nodes['oral'].replace(0, False)
nodes['parenteral'] = nodes['parenteral'].replace(0, False)
nodes['topical'] = nodes['topical'].replace(0, False)

#show value count for oral, parenteral and topical
print(nodes['oral'].value_counts())
print(nodes['parenteral'].value_counts())
print(nodes['topical'].value_counts())

True     981
False    958
Name: oral, dtype: int64
False    1274
True      665
Name: parenteral, dtype: int64
False    1657
True      282
Name: topical, dtype: int64

5. Basic Graph Metrics

Using the weights (minor interaction = 3, major = 1, unknown = 2, medium = 2) for centrality measures in a social network analysis of drug interactions can make sense if you want to focus on the severe drugs and avoid these. All further analysis focus only on subgraph 1.

Emphasizing major interactions: Assigning a lower weight (1) to major interactions means that major interactions will have a stronger influence on the centrality measures. This is because, in many centrality calculations, lower weights are interpreted as stronger connections. By prioritizing major interactions in this way, the analysis can focus on understanding and addressing the most severe drug interactions.

Treating unknown and medium interactions similarly: Assigning the same weight (2) to both unknown and medium interactions indicates that the analysis considers them to have similar importance. This is a conservative approach that assumes that the severity of an unknown interaction could be similar to that of a medium interaction. By using this weighting, the analysis can avoid underestimating the potential risks of unknown interactions.

Penalizing minor interactions: Assigning a higher weight (3) to minor interactions implies that these interactions will have a weaker influence on the centrality measures. In many contexts, minor interactions may not be as concerning as major or medium interactions. By giving them a higher weight, the analysis can focus more on the more severe interactions that may pose a greater risk to patients.

Use cases where this weighting scheme might be beneficial include:

• Identifying drugs with high potential for severe interactions: Healthcare providers and researchers can use these centrality measures to identify drugs that are more likely to have major interactions with other drugs. This information can guide decision-making in prescribing medications and help prevent adverse drug events.

• Drug development and clinical trial design: Pharmaceutical companies can use these weights to understand potential drug interactions and identify areas for improvement during drug development. This knowledge can also help in designing clinical trials, as researchers can be more cautious when combining drugs with high potential for major interactions.

• Pharmacovigilance and drug safety monitoring: Regulatory agencies can use these centrality measures to monitor drug safety, prioritize drugs for further investigation, and develop strategies for managing potential drug interactions.

In summary, using the specified weights for centrality measures can help prioritize major drug interactions and handle unknown interactions conservatively, making it useful in various healthcare and pharmaceutical contexts. However, it is essential to consider the specific research question and context when choosing weights, as other weighting schemes might be more appropriate depending on the goals of the analysis.

# calculate and add closeness centrality to nodes dataframe
closeness_centrality = nx.closeness_centrality(G)
nodes['closeness_centrality'] = nodes['DDInterID'].apply(lambda x: closeness_centrality[x])

# calculate and add weighted closeness centrality to nodes dataframe
weighted_closeness_centrality = nx.closeness_centrality(G, distance='Weight')
nodes['weighted_closeness_centrality'] = nodes['DDInterID'].apply(lambda x: weighted_closeness_centrality[x])


# calculate and add betweenness centrality to nodes dataframe
betweenness_centrality = nx.betweenness_centrality(G)
nodes['betweenness_centrality'] = nodes['DDInterID'].apply(lambda x: betweenness_centrality[x])

# calculate and add weighted betweenness centrality to nodes dataframe
weighted_betweenness_centrality = nx.betweenness_centrality(G, weight='Weight')
nodes['weighted_betweenness_centrality'] = nodes['DDInterID'].apply(lambda x: weighted_betweenness_centrality[x])

# calculate and add PageRank to nodes dataframe
pagerank = nx.pagerank(G)
nodes['pagerank'] = nodes['DDInterID'].apply(lambda x: pagerank[x])

# calculate and add weighted PageRank to nodes dataframe
weighted_pagerank = nx.pagerank(G, weight='Weight')
nodes['weighted_pagerank'] = nodes['DDInterID'].apply(lambda x: weighted_pagerank[x])

# calculate and add degree centrality to nodes dataframe
degree_centrality = nx.degree_centrality(G)
nodes['degree_centrality'] = nodes['DDInterID'].apply(lambda x: betweenness_centrality[x])

# calculate and add degree to nodes dataframe
degree = nx.degree(G)
nodes['degree'] = nodes['DDInterID'].apply(lambda x: degree[x])


#calculate and add eigenvector centrality to nodes dataframe
eigenvector_centrality = nx.eigenvector_centrality(G)
nodes['eigenvector_centrality'] = nodes['DDInterID'].apply(lambda x: eigenvector_centrality[x])

#calculate and add weighted eigenvector centrality to nodes dataframe
weighted_eigenvector_centrality = nx.eigenvector_centrality(G, weight='Weight')
nodes['weighted_eigenvector_centrality'] = nodes['DDInterID'].apply(lambda x: weighted_eigenvector_centrality[x])


#calculate and add clustering coefficient to nodes dataframe
clustering_coefficient = nx.clustering(G)
nodes['clustering_coefficient'] = nodes['DDInterID'].apply(lambda x: clustering_coefficient[x])

#calculate and add weighted clustering coefficient to nodes dataframe
weighted_clustering_coefficient = nx.clustering(G, weight='Weight')
nodes['weighted_clustering_coefficient'] = nodes['DDInterID'].apply(lambda x: weighted_clustering_coefficient[x])

nodes.head(2)

	DDInterID	Drug	Drug_original	Units MAT Feb 2023	% Growth	Sum incl combi	Ranking Mono	Ranking Combi	topical	nasal	ophthalmic	oral	parenteral	indication_class	chirality	black_box_warning	availability_type	first_in_class	ATC_Classifications	ATC1_names	ANTIINFECTIVES_FOR_SYSTEMIC_USE	RESPIRATORY_SYSTEM	ANTIPARASITIC_PRODUCTS,_INSECTICIDES_AND_REPELLENTS	ANTINEOPLASTIC_AND_IMMUNOMODULATING_AGENTS	MUSCULO-SKELETAL_SYSTEM	ALIMENTARY_TRACT_AND_METABOLISM	BLOOD_AND_BLOOD_FORMING_ORGANS	SENSORY_ORGANS	SYSTEMIC_HORMONAL_PREPARATIONS,_EXCL._SEX_HORMONES_AND_INSULINS	NERVOUS_SYSTEM	GENITO_URINARY_SYSTEM_AND_SEX_HORMONES	DERMATOLOGICALS	VARIOUS	CARDIOVASCULAR_SYSTEM	ATC2_names	R01_-_NASAL_PREPARATIONS	A01_-_STOMATOLOGICAL_PREPARATIONS	S02_-_OTOLOGICALS	S03_-_OPHTHALMOLOGICAL_AND_OTOLOGICAL_PREPARATIONS	D07_-_CORTICOSTEROIDS,_DERMATOLOGICAL_PREPARATIONS	D10_-_ANTI-ACNE_PREPARATIONS	S01_-_OPHTHALMOLOGICALS	C05_-_VASOPROTECTIVES	H02_-_CORTICOSTEROIDS_FOR_SYSTEMIC_USE	R07_-_OTHER_RESPIRATORY_SYSTEM_PRODUCTS	J05_-_ANTIVIRALS_FOR_SYSTEMIC_USE	R03_-_DRUGS_FOR_OBSTRUCTIVE_AIRWAY_DISEASES	A07_-_ANTIDIARRHEALS,_INTESTINAL_ANTIINFLAMMATORY/ANTIINFECTIVE_AGENTS	R06_-_ANTIHISTAMINES_FOR_SYSTEMIC_USE	N06_-_PSYCHOANALEPTICS	D11_-_OTHER_DERMATOLOGICAL_PREPARATIONS	N02_-_ANALGESICS	R05_-_COUGH_AND_COLD_PREPARATIONS	N07_-_OTHER_NERVOUS_SYSTEM_DRUGS	P03_-_ECTOPARASITICIDES,_INCL._SCABICIDES,_INSECTICIDES_AND_REPELLENTS	R02_-_THROAT_PREPARATIONS	D04_-_ANTIPRURITICS,_INCL._ANTIHISTAMINES,_ANESTHETICS,_ETC.	N01_-_ANESTHETICS	C01_-_CARDIAC_THERAPY	C04_-_PERIPHERAL_VASODILATORS	L04_-_IMMUNOSUPPRESSANTS	B01_-_ANTITHROMBOTIC_AGENTS	L02_-_ENDOCRINE_THERAPY	L01_-_ANTINEOPLASTIC_AGENTS	A11_-_VITAMINS	M02_-_TOPICAL_PRODUCTS_FOR_JOINT_AND_MUSCULAR_PAIN	M01_-_ANTIINFLAMMATORY_AND_ANTIRHEUMATIC_PRODUCTS	G02_-_OTHER_GYNECOLOGICALS	A10_-_DRUGS_USED_IN_DIABETES	C07_-_BETA_BLOCKING_AGENTS	J02_-_ANTIMYCOTICS_FOR_SYSTEMIC_USE	D01_-_ANTIFUNGALS_FOR_DERMATOLOGICAL_USE	N05_-_PSYCHOLEPTICS	H01_-_PITUITARY_AND_HYPOTHALAMIC_HORMONES_AND_ANALOGUES	N03_-_ANTIEPILEPTICS	M03_-_MUSCLE_RELAXANTS	N04_-_ANTI-PARKINSON_DRUGS	G04_-_UROLOGICALS	A03_-_DRUGS_FOR_FUNCTIONAL_GASTROINTESTINAL_DISORDERS	C02_-_ANTIHYPERTENSIVES	A04_-_ANTIEMETICS_AND_ANTINAUSEANTS	D08_-_ANTISEPTICS_AND_DISINFECTANTS	V03_-_ALL_OTHER_THERAPEUTIC_PRODUCTS	A08_-_ANTIOBESITY_PREPARATIONS,_EXCL._DIET_PRODUCTS	A12_-_MINERAL_SUPPLEMENTS	B05_-_BLOOD_SUBSTITUTES_AND_PERFUSION_SOLUTIONS	B02_-_ANTIHEMORRHAGICS	L03_-_IMMUNOSTIMULANTS	C03_-_DIURETICS	C09_-_AGENTS_ACTING_ON_THE_RENIN-ANGIOTENSIN_SYSTEM	A02_-_DRUGS_FOR_ACID_RELATED_DISORDERS	J01_-_ANTIBACTERIALS_FOR_SYSTEMIC_USE	D06_-_ANTIBIOTICS_AND_CHEMOTHERAPEUTICS_FOR_DERMATOLOGICAL_USE	A06_-_DRUGS_FOR_CONSTIPATION	P01_-_ANTIPROTOZOALS	J04_-_ANTIMYCOBACTERIALS	V04_-_DIAGNOSTIC_AGENTS	G03_-_SEX_HORMONES_AND_MODULATORS_OF_THE_GENITAL_SYSTEM	M09_-_OTHER_DRUGS_FOR_DISORDERS_OF_THE_MUSCULO-SKELETAL_SYSTEM	G01_-_GYNECOLOGICAL_ANTIINFECTIVES_AND_ANTISEPTICS	C08_-_CALCIUM_CHANNEL_BLOCKERS	M05_-_DRUGS_FOR_TREATMENT_OF_BONE_DISEASES	V08_-_CONTRAST_MEDIA	C10_-_LIPID_MODIFYING_AGENTS	V10_-_THERAPEUTIC_RADIOPHARMACEUTICALS	D05_-_ANTIPSORIATICS	A16_-_OTHER_ALIMENTARY_TRACT_AND_METABOLISM_PRODUCTS	M04_-_ANTIGOUT_PREPARATIONS	B06_-_OTHER_HEMATOLOGICAL_AGENTS	A14_-_ANABOLIC_AGENTS_FOR_SYSTEMIC_USE	P02_-_ANTHELMINTICS	H05_-_CALCIUM_HOMEOSTASIS	A05_-_BILE_AND_LIVER_THERAPY	V09_-_DIAGNOSTIC_RADIOPHARMACEUTICALS	H03_-_THYROID_THERAPY	B03_-_ANTIANEMIC_PREPARATIONS	D09_-_MEDICATED_DRESSINGS	D03_-_PREPARATIONS_FOR_TREATMENT_OF_WOUNDS_AND_ULCERS	D02_-_EMOLLIENTS_AND_PROTECTIVES	J07_-_VACCINES	A09_-_DIGESTIVES,_INCL._ENZYMES	V06_-_GENERAL_NUTRIENTS	H04_-_PANCREATIC_HORMONES	subgraph	closeness_centrality	weighted_closeness_centrality	betweenness_centrality	weighted_betweenness_centrality	pagerank	weighted_pagerank	degree_centrality	degree	eigenvector_centrality	weighted_eigenvector_centrality	clustering_coefficient	weighted_clustering_coefficient
0	DDInter513	Dexamethasone	Dexamethasone	401082.0	0.055232	1162374.0	75.0	44.0	True	False	False	True	True	Glucocorticoid	1.0	0.0	1.0	0.0	R01AD53, A01AC02, R01AD03, S02BA06, S03BA01, D...	{'RESPIRATORY SYSTEM': 2, 'ALIMENTARY TRACT AN...	0	2	0	0	0	1	0	4	1	0	0	3	0	1	{'R01 - NASAL PREPARATIONS': 1, 'A01 - STOMATO...	1	1	1	1	1	1	1	1	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.635260	0.304632	0.011539	0.011286	0.002508	0.002468	0.011539	910	0.077820	0.075080	0.205962	0.148061
1	DDInter997	Ivacaftor	Ivacaftor	0.0	0.000000	0.0	0.0	0.0	False	False	False	True	0	0	2.0	0.0	1.0	1.0	R07AX02	{'RESPIRATORY SYSTEM': 1}	0	1	0	0	0	0	0	0	0	0	0	0	0	0	{'R07 - OTHER RESPIRATORY SYSTEM PRODUCTS': 1}	0	0	0	0	0	0	0	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.527212	0.255197	0.000709	0.000842	0.001023	0.001022	0.000709	377	0.037246	0.038606	0.351685	0.262743

5.1 Closeness centrality

Interpretation: A drug with high closeness centrality can quickly influence or be influenced by other drugs in the network due to its short average distance to all other drugs. HCP Action: HCPs should be aware that a drug with high closeness centrality may have a more significant overall impact on the patient's medication regimen. They should evaluate the entire drug profile of the patient to identify and manage potential interactions.

# Calculate the closeness centrality for each node
cc = nx.closeness_centrality(G)

# Create a histogram of the closeness centrality values
fig = go.Figure(data=[go.Histogram(x=list(cc.values()))])

# Add labels and title to the plot
fig.update_layout(
    title="Closeness Centrality Distribution",
    xaxis_title="Closeness Centrality",
    yaxis_title="Count"
)

# Show the plot
fig.show()

# Filter by subgraph 1 and sort by Top 20 closeness centrality, show columns closeness_centrality, Drug_original, Ranking Mono,Ranking Mono,oral,parenteral,indication_class,topical,chirality,availability_type,black_box_warning,first_in_class	ATC_Classification
nodes[nodes['subgraph'] == 1].sort_values('closeness_centrality', ascending=False)[['closeness_centrality', 'Drug_original', 'oral','Ranking Mono ','Ranking Combi', 'parenteral', 'indication_class', 'topical', 'chirality', 'availability_type', 'black_box_warning', 'first_in_class','ATC1_names']].head(10)

	closeness_centrality	Drug_original	oral	Ranking Mono	Ranking Combi	parenteral	indication_class	topical	chirality	availability_type	black_box_warning	first_in_class	ATC1_names
0	0.635260	Dexamethasone	True	75.0	44.0	True	Glucocorticoid	True	1.0	1.0	0.0	0.0	{'RESPIRATORY SYSTEM': 2, 'ALIMENTARY TRACT AN...
417	0.626385	Tacrolimus	True	159.0	248.0	True	Immunosuppressant	True	1.0	1.0	1.0	0.0	{}
580	0.624761	Warfarin	True	0.0	0.0	True	Anticoagulant	False	0.0	1.0	1.0	0.0	{'BLOOD AND BLOOD FORMING ORGANS': 1}
37	0.618151	Promethazine	True	0.0	0.0	True	Anti-Emetic, Antihistaminic	True	0.0	1.0	1.0	0.0	{'RESPIRATORY SYSTEM': 2, 'DERMATOLOGICALS': 1}
409	0.613626	Prednisone	True	85.0	157.0	False	Glucocorticoid	False	1.0	1.0	0.0	0.0	{'ALIMENTARY TRACT AND METABOLISM': 1, 'SYSTEM...
698	0.613431	Clarithromycin	True	214.0	307.0	False	Antibacterial	False	1.0	1.0	0.0	0.0	{'ANTIINFECTIVES FOR SYSTEMIC USE': 1}
484	0.613041	Cyclosporine	True	0.0	0.0	True	Immunosuppressant	True	1.0	1.0	1.0	0.0	{'SENSORY ORGANS': 1, 'ANTINEOPLASTIC AND IMMU...
239	0.611679	Thalidomide	True	0.0	0.0	False	Sedative-Hypnotic	False	0.0	1.0	1.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}
529	0.610710	Methotrexate	False	31.0	80.0	False	Antineoplastic	False	-1.0	-1.0	0.0	-1.0	{}
74	0.607439	Fluconazole	True	107.0	188.0	True	Antifungal	False	2.0	1.0	0.0	0.0	{'ANTIINFECTIVES FOR SYSTEMIC USE': 1, 'DERMAT...

The closeness centrality and weighted closeness centrality of various drugs in a network built based on drug interactions. Closeness centrality is a measure of how central a node (drug) is within the network, indicating how easily it can be reached from other nodes (drugs) in the network. The higher the closeness centrality, the more central the drug is in the network. Weighted closeness centrality takes into account the weights (interaction strength) between nodes, with lower weights indicating worse interactions.

Unweighted:

• Dexamethasone a glucocorticoid, has the highest closeness centrality (0.635260), indicating that it is the most central drug in both unweighted networks. It is commonly used to treat a variety of inflammatory and autoimmune conditions and is also used in respiratory and alimentary tract-related treatments.

• Tacrolimus an immunosuppressant, has the second-highest closeness centrality (0.626385). It is mainly used to prevent organ transplant rejection and is available in both oral and parenteral forms.

• Warfarin an anticoagulant, has the third-highest closeness centrality (0.624761) and the fourth-highest weighted closeness centrality (0.297809). It is used to prevent blood clots and is available in oral and parenteral forms.

• Promethazine an antihistaminic and anti-emetic, ranks fourth in closeness centrality (0.618151) and second in weighted closeness centrality (0.300407). It is used for treating allergic reactions, nausea, and vomiting.

Weighted:

• Ozanimod has the highest weighted closeness centrality score, which means it is the most central drug in the network when considering the interaction severity (weights). This implies that Ozanimod may have more connections to other drugs, especially those with more severe interactions, making it an important drug to consider when studying drug interactions.

• Siponimod is a selective sphingosine-1-phosphate (S1P) receptor modulator, primarily used for the treatment of relapsing multiple sclerosis. Its high weighted closeness centrality indicates that it has many connections to other drugs, particularly those with more severe interactions. This suggests that Siponimod may be an important drug to consider when examining drug interactions in patients with multiple sclerosis, as it might require careful monitoring and management to avoid adverse effects.

• Leflunomide is a disease-modifying antirheumatic drug (DMARD) used to treat rheumatoid arthritis and psoriatic arthritis. The high weighted closeness centrality score suggests that Leflunomide has numerous connections with other drugs, especially those with more severe interactions. This indicates that Leflunomide could play a crucial role in drug interactions for patients with arthritis and might require special attention to prevent adverse effects when used in combination with other medications.

• Teriflunomide Teriflunomide is an immunomodulatory drug used to treat relapsing forms of multiple sclerosis. Its high weighted closeness centrality implies that it has many connections to other drugs, particularly those with more severe interactions. This suggests that Teriflunomide is an important drug to consider in the context of drug interactions in multiple sclerosis treatment. Careful management and monitoring may be needed when Teriflunomide is used with other medications to minimize the risk of adverse effects.

General Insights

Glucocorticoids (Dexamethasone, Prednisone, Prednisolone, and Hydrocortisone) are well-represented in the top ranking molecules by closeness centrality, indicating their importance and widespread use due to the top ranking in "Mono and Ranking in Combi" in treating various conditions.

Immunosuppressants (Tacrolimus and Cyclosporine) and antibiotics (Clarithromycin) are not widley used as top molecules, but with the interaction highlighting their significance in treating infections and managing immune responses.

The data also includes drugs for specific conditions, such as Warfarin for blood clotting,Promethazine as Anti-Emetic, Antihistaminic, and Tacrolimus for Immunosuppressant are drugs where the FDA's already put the most stringent warning for the drug. These drugs have a relatively high centrality in the network, suggesting that they have interactions with multiple other drugs, which gives health care providers warning of serious side effects of injury or death.

# Filter by subgraph 1 and sort by Top 20 closeness centrality, show columns closeness_centrality, Drug_original, Ranking Mono,Ranking Mono,oral,parenteral,indication_class,topical,chirality,availability_type,black_box_warning,first_in_class	ATC_Classification
nodes[nodes['subgraph'] == 1].sort_values('weighted_closeness_centrality', ascending=False)[['weighted_closeness_centrality', 'Drug_original', 'Ranking Mono ','Ranking Combi', 'oral','parenteral', 'indication_class', 'topical', 'chirality', 'availability_type', 'black_box_warning', 'first_in_class', 'ATC1_names']].head(10)

	weighted_closeness_centrality	Drug_original	Ranking Mono	Ranking Combi	oral	parenteral	indication_class	topical	chirality	availability_type	black_box_warning	first_in_class	ATC1_names
1508	0.462353	Ozanimod	831.0	895.0	True	False	0	False	1.0	1.0	0.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}
1510	0.459707	Siponimod	930.0	981.0	True	False	0	False	2.0	1.0	0.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}
756	0.457852	Leflunomide	480.0	583.0	True	False	0	False	2.0	1.0	1.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}
1088	0.451314	Teriflunomide	824.0	886.0	True	False	0	False	2.0	1.0	1.0	1.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}
1022	0.444345	Fingolimod	646.0	728.0	True	False	0	False	2.0	1.0	0.0	1.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}
697	0.436795	Ceritinib	1177.0	1206.0	True	False	0	False	2.0	1.0	0.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}
914	0.432193	Panobinostat	1220.0	1240.0	True	False	0	False	2.0	1.0	1.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}
123	0.431226	Clozapine	178.0	267.0	True	False	Antipsychotic	False	2.0	1.0	1.0	0.0	{'NERVOUS SYSTEM': 1}
417	0.430072	Tacrolimus	159.0	248.0	True	True	Immunosuppressant	True	1.0	1.0	1.0	0.0	{}
713	0.426646	Ribociclib	785.0	853.0	True	False	0	False	2.0	1.0	0.0	1.0	{}

5.2 Betweeness Centrality

Interpretation: A drug with high betweenness centrality often lies on the shortest path between other drug pairs, acting as a bridge or mediator in the network. HCP Action: HCPs should recognize that a drug with high betweenness centrality may play a crucial role in connecting different drug communities. Prescribing or removing such a drug may have a significant impact on the patient's overall medication profile, and HCPs should closely monitor for changes in drug interactions.

# Calculate the closeness centrality for each node
bb = nx.betweenness_centrality(G)

# Create a histogram of the closeness centrality values
fig = go.Figure(data=[go.Histogram(x=list(bb.values()))])

# Add labels and title to the plot
fig.update_layout(
    title="Betweeness Centrality Distribution",
    xaxis_title="BetweenessCentrality",
    yaxis_title="Count"
)

# Show the plot
fig.show()

Betweenness Centrality:

• Warfarin (Anticoagulant) has the highest betweenness centrality value (0.025057) among the listed drugs. This suggests that Warfarin is critical in connecting different drugs in the network, possibly indicating that it has multiple interactions with other drugs, and thus may require close monitoring for potential drug-drug interactions.

• Ozanimod, Idelalisib, and Thalidomide also have relatively high betweenness centrality values. These drugs are classified under 'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS', suggesting that they might play significant roles in connecting drugs within this therapeutic category.

• Methotrexate , which is an Antineoplastic drug, has a relatively high betweenness centrality value (0.012522), implying that it might have many connections with other drugs within the network.

• Cyclosporine (Immunosuppressant) and Dexamethasone (Glucocorticoid) also have relatively high betweenness centrality values, suggesting that they are crucial in connecting different drugs in the network, possibly due to their multiple therapeutic uses and drug interactions.

Some drugs with relatively lower betweenness centrality values, such as Picosulfuric acid and Carfilzomib, might have fewer connections with other drugs in the network, indicating that they may have fewer drug interactions.

In conclusion, betweenness centrality can provide insights into the importance of different drugs in the network, potentially highlighting drugs with multiple interactions that may require close monitoring for potential drug-drug interactions. However, it is essential to note that the provided data is based on unweighted betweenness centrality, which may not fully account for the severity of the interactions. Incorporating weights based on interaction severity could provide a more accurate representation of the drug interaction network.

# Filter by subgraph 1 and sort by Top 15 closeness centrality, show columns closeness_centrality, Drug_original, Ranking Mono,Ranking Mono,oral,parenteral,indication_class,topical,chirality,availability_type,black_box_warning,first_in_class	ATC_Classification
nodes[nodes['subgraph'] == 1].sort_values('betweenness_centrality', ascending=False)[['betweenness_centrality', 'Drug_original', 'Ranking Mono ','Ranking Combi', 'oral','parenteral', 'indication_class', 'topical', 'chirality', 'availability_type', 'black_box_warning', 'first_in_class', 'ATC1_names','ATC_Classifications']].head(10)

	betweenness_centrality	Drug_original	Ranking Mono	Ranking Combi	oral	parenteral	indication_class	topical	chirality	availability_type	black_box_warning	first_in_class	ATC1_names	ATC_Classifications
580	0.025057	Warfarin	0.0	0.0	True	True	Anticoagulant	False	0.0	1.0	1.0	0.0	{'BLOOD AND BLOOD FORMING ORGANS': 1}	B01AA03
1508	0.015335	Ozanimod	831.0	895.0	True	False	0	False	1.0	1.0	0.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}	L04AA38
705	0.012598	Idelalisib	1158.0	1187.0	True	False	0	False	1.0	1.0	1.0	1.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}	L01EM01
239	0.012547	Thalidomide	0.0	0.0	True	False	Sedative-Hypnotic	False	0.0	1.0	1.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}	L04AX02
529	0.012522	Methotrexate	31.0	80.0	False	False	Antineoplastic	False	-1.0	-1.0	0.0	-1.0	{}	0
484	0.011952	Cyclosporine	0.0	0.0	True	True	Immunosuppressant	True	1.0	1.0	1.0	0.0	{'SENSORY ORGANS': 1, 'ANTINEOPLASTIC AND IMMU...	S01XA18, L04AD01
0	0.011539	Dexamethasone	75.0	44.0	True	True	Glucocorticoid	True	1.0	1.0	0.0	0.0	{'RESPIRATORY SYSTEM': 2, 'ALIMENTARY TRACT AN...	R01AD53, A01AC02, R01AD03, S02BA06, S03BA01, D...
1280	0.010379	Picosulfuric acid	0.0	0.0	False	False	0	False	0.0	0.0	0.0	0.0	{}	0
657	0.009685	Chloramphenicol	0.0	0.0	True	True	Antibacterial, Antirickettsial	True	1.0	-1.0	0.0	0.0	{'SENSORY ORGANS': 3, 'DERMATOLOGICALS': 2, 'G...	S01AA01, S03AA08, D06AX02, G01AA05, S02AA01, D...
37	0.009564	Promethazine	0.0	0.0	True	True	Anti-Emetic, Antihistaminic	True	0.0	1.0	1.0	0.0	{'RESPIRATORY SYSTEM': 2, 'DERMATOLOGICALS': 1}	R06AD52, D04AA10, R06AD02

Considering the weighted betweenness centrality along with the unweighted betweenness centrality and other drug characteristics, here are some conclusions about the drug interactions:

Ozanimod see previous section

Siponimod is an immunomodulating agent used for the treatment of multiple sclerosis. Its high unweighted betweenness centrality indicates that it has numerous connections with other drugs. This suggests that Siponimod may be an important drug to consider in the context of drug interactions for multiple sclerosis treatment, and careful monitoring and management may be needed when it is used with other medications.

Leflunomide is a disease-modifying antirheumatic drug (DMARD) used to treat rheumatoid arthritis and psoriatic arthritis. Its high unweighted betweenness centrality suggests that it has many connections with other drugs. This indicates that Leflunomide could play a crucial role in drug interactions for patients with arthritis and might require special attention to prevent adverse effects when used in combination with other medications.

Conclusion: Drugs with high betweenness centrality scores, such as Warfarin, Ozanimod, Idelalisib, Siponimod, and Leflunomide, are likely to have numerous connections with other drugs and play a central role in drug interactions. These drugs are used in the treatment of various conditions, including anticoagulation, autoimmune disorders, and cancer. Healthcare providers should carefully monitor and manage these medications when used with other drugs to minimize the risk of adverse effects and ensure the safety and effectiveness of treatment.

# Filter by subgraph 1 and sort by Top 15 weighted betweeness, show columns closeness_centrality, Drug_original, Ranking Mono,Ranking Mono,oral,parenteral,indication_class,topical,chirality,availability_type,black_box_warning,first_in_class	ATC_Classification
nodes[nodes['subgraph'] == 1].sort_values('weighted_betweenness_centrality', ascending=False)[['weighted_betweenness_centrality', 'Drug_original', 'Ranking Mono ','Ranking Combi', 'oral','parenteral', 'indication_class', 'topical', 'chirality', 'availability_type', 'black_box_warning', 'first_in_class', 'ATC1_names','ATC_Classifications']].head(10)

	weighted_betweenness_centrality	Drug_original	Ranking Mono	Ranking Combi	oral	parenteral	indication_class	topical	chirality	availability_type	black_box_warning	first_in_class	ATC1_names	ATC_Classifications
1508	0.052146	Ozanimod	831.0	895.0	True	False	0	False	1.0	1.0	0.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}	L04AA38
1510	0.048535	Siponimod	930.0	981.0	True	False	0	False	2.0	1.0	0.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}	L04AA42
756	0.047223	Leflunomide	480.0	583.0	True	False	0	False	2.0	1.0	1.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}	L04AA13
1088	0.045521	Teriflunomide	824.0	886.0	True	False	0	False	2.0	1.0	1.0	1.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}	L04AA31
13	0.021420	Hydrocodone	1337.0	1328.0	False	False	Antitussive	False	1.0	-1.0	0.0	0.0	{'RESPIRATORY SYSTEM': 1}	R05DA03
697	0.021399	Ceritinib	1177.0	1206.0	True	False	0	False	2.0	1.0	0.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}	L01ED02
239	0.020314	Thalidomide	0.0	0.0	True	False	Sedative-Hypnotic	False	0.0	1.0	1.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}	L04AX02
580	0.019791	Warfarin	0.0	0.0	True	True	Anticoagulant	False	0.0	1.0	1.0	0.0	{'BLOOD AND BLOOD FORMING ORGANS': 1}	B01AA03
529	0.018436	Methotrexate	31.0	80.0	False	False	Antineoplastic	False	-1.0	-1.0	0.0	-1.0	{}	0
1397	0.018347	Iopamidol	663.0	742.0	False	True	Diagnostic Aid (radiopaque medium)	False	1.0	1.0	0.0	0.0	{'VARIOUS': 1}	V08AB04

5.3 Pagerank

Interpretation: A drug with a high PageRank score is considered important if it is connected to other important drugs in the network. HCP Action: HCPs should consider the influence a high PageRank drug may have on other significant drugs in the network. They should evaluate the patient's entire medication regimen and may need to adjust dosages or select alternative treatments to minimize the risk of adverse drug interactions.

Pagerank Top 3:

• Warfarin - Anticoagulant, both oral and parenteral, with a black box warning and first in class.
• Dexamethasone - Glucocorticoid, both oral and parenteral, with no black box warning and not first in class.
• Cyclosporine - Immunosuppressant, both oral and parenteral, with a black box warning and first in class.

# Filter by subgraph 1 and sort by Top 15 closeness centrality, show columns closeness_centrality, Drug_original, Ranking Mono,Ranking Mono,oral,parenteral,indication_class,topical,chirality,availability_type,black_box_warning,first_in_class	ATC_Classification
nodes[nodes['subgraph'] == 1].sort_values('pagerank', ascending=False)[['pagerank', 'Drug_original', 'Ranking Mono ','Ranking Combi', 'oral','parenteral', 'indication_class', 'topical', 'chirality', 'availability_type', 'black_box_warning', 'first_in_class', 'ATC1_names']].head(10)

	pagerank	Drug_original	Ranking Mono	Ranking Combi	oral	parenteral	indication_class	topical	chirality	availability_type	black_box_warning	first_in_class	ATC1_names
580	0.002886	Warfarin	0.0	0.0	True	True	Anticoagulant	False	0.0	1.0	1.0	0.0	{'BLOOD AND BLOOD FORMING ORGANS': 1}
0	0.002508	Dexamethasone	75.0	44.0	True	True	Glucocorticoid	True	1.0	1.0	0.0	0.0	{'RESPIRATORY SYSTEM': 2, 'ALIMENTARY TRACT AN...
484	0.002397	Cyclosporine	0.0	0.0	True	True	Immunosuppressant	True	1.0	1.0	1.0	0.0	{'SENSORY ORGANS': 1, 'ANTINEOPLASTIC AND IMMU...
417	0.002313	Tacrolimus	159.0	248.0	True	True	Immunosuppressant	True	1.0	1.0	1.0	0.0	{}
37	0.002269	Promethazine	0.0	0.0	True	True	Anti-Emetic, Antihistaminic	True	0.0	1.0	1.0	0.0	{'RESPIRATORY SYSTEM': 2, 'DERMATOLOGICALS': 1}
529	0.002246	Methotrexate	31.0	80.0	False	False	Antineoplastic	False	-1.0	-1.0	0.0	-1.0	{}
698	0.002221	Clarithromycin	214.0	307.0	True	False	Antibacterial	False	1.0	1.0	0.0	0.0	{'ANTIINFECTIVES FOR SYSTEMIC USE': 1}
409	0.002221	Prednisone	85.0	157.0	True	False	Glucocorticoid	False	1.0	1.0	0.0	0.0	{'ALIMENTARY TRACT AND METABOLISM': 1, 'SYSTEM...
239	0.002171	Thalidomide	0.0	0.0	True	False	Sedative-Hypnotic	False	0.0	1.0	1.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}
705	0.002153	Idelalisib	1158.0	1187.0	True	False	0	False	1.0	1.0	1.0	1.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}

Weighted Pagerank Top 3:

• Warfarin - Anticoagulant, both oral and parenteral, with a black box warning and first in class.
• Dexamethasone - Glucocorticoid, both oral and parenteral, with no black box warning and not first in class.
• Ozanimod - Part of the 'Antineoplastic and Immunomodulating Agents' category, oral only, with a black box warning and not first in class.

# Filter by subgraph 1 and sort by Top 15 weighted_pagerank, show columns closeness_centrality, Drug_original, Ranking Mono,Ranking Mono,oral,parenteral,indication_class,topical,chirality,availability_type,black_box_warning,first_in_class	ATC_Classification
nodes[nodes['subgraph'] == 1].sort_values('weighted_pagerank', ascending=False)[['weighted_pagerank', 'Drug_original', 'Ranking Mono ','Ranking Combi', 'oral','parenteral', 'indication_class', 'topical', 'chirality', 'availability_type', 'black_box_warning', 'first_in_class', 'ATC1_names']].head(10)

	weighted_pagerank	Drug_original	Ranking Mono	Ranking Combi	oral	parenteral	indication_class	topical	chirality	availability_type	black_box_warning	first_in_class	ATC1_names
580	0.002933	Warfarin	0.0	0.0	True	True	Anticoagulant	False	0.0	1.0	1.0	0.0	{'BLOOD AND BLOOD FORMING ORGANS': 1}
0	0.002574	Dexamethasone	75.0	44.0	True	True	Glucocorticoid	True	1.0	1.0	0.0	0.0	{'RESPIRATORY SYSTEM': 2, 'ALIMENTARY TRACT AN...
484	0.002420	Cyclosporine	0.0	0.0	True	True	Immunosuppressant	True	1.0	1.0	1.0	0.0	{'SENSORY ORGANS': 1, 'ANTINEOPLASTIC AND IMMU...
409	0.002327	Prednisone	85.0	157.0	True	False	Glucocorticoid	False	1.0	1.0	0.0	0.0	{'ALIMENTARY TRACT AND METABOLISM': 1, 'SYSTEM...
37	0.002323	Promethazine	0.0	0.0	True	True	Anti-Emetic, Antihistaminic	True	0.0	1.0	1.0	0.0	{'RESPIRATORY SYSTEM': 2, 'DERMATOLOGICALS': 1}
529	0.002241	Methotrexate	31.0	80.0	False	False	Antineoplastic	False	-1.0	-1.0	0.0	-1.0	{}
28	0.002190	Salbutamol	32.0	72.0	True	False	Bronchodilator	True	0.0	1.0	1.0	0.0	{'RESPIRATORY SYSTEM': 2}
417	0.002183	Tacrolimus	159.0	248.0	True	True	Immunosuppressant	True	1.0	1.0	1.0	0.0	{}
20	0.002171	Prednisolone	53.0	58.0	True	False	Glucocorticoid	True	1.0	1.0	0.0	0.0	{'SENSORY ORGANS': 4, 'DERMATOLOGICALS': 2, 'R...
698	0.002157	Clarithromycin	214.0	307.0	True	False	Antibacterial	False	1.0	1.0	0.0	0.0	{'ANTIINFECTIVES FOR SYSTEMIC USE': 1}

From the analysis, Warfarin consistently ranks highest in all three measures. It is an anticoagulant that is available both orally and parenterally, with a black box warning and is first in class. Dexamethasone ranks high in pagerank and weighted pagerank, and it is a glucocorticoid available both orally and parenterally with no black box warning and not first in class. Ozanimod appears in the top 3 of betweenness centrality and weighted pagerank and belongs to the 'Antineoplastic and Immunomodulating Agents' category, being orally available, with a black box warning, and not first in class.

It's important to consider both mono and combi rankings while analyzing these drugs. For instance, Methotrexate has a relatively high mono ranking (31) and combi ranking (71), while Warfarin seems not be used in both mono and combi. This suggests that although Warfarin ranks high in network-based metrics, it seems not be widely used compared to Methotrexate.

5.4. Degree and degree Centrality

In this analysis, we focus on the differences in degree centrality and degree for the given data. Degree centrality is a measure of the importance of a node (drug) in a network, and the degree of a node is the number of connections it has to other nodes.

Key Points:

• Warfarin has the highest degree centrality (0.025057) among the drugs, followed by Ozanimod (0.015335) and Idelalisib (0.012598). This suggests that these drugs have a greater influence or importance in the network compared to others.

• Dexamethasone has the highest degree (910), indicating that it has the most connections with other drugs. Other drugs with high degrees include Tacrolimus (866), Warfarin (864), Promethazine (832), and Clarithromycin (822).

Interestingly, Warfarin has both high degree centrality and degree, suggesting that it has many connections and is a crucial drug in the network.

Some drugs have a significant difference in their ranking for mono and combi therapy, such as Tacrolimus (159 for mono and 243 for combi) and Ozanimod (831 for mono and 1123 for combi). This could indicate that these drugs have different levels of effectiveness when used alone or in combination with other drugs.

A majority of the drugs in the dataset are orally administered, with some also available for parenteral use, like Dexamethasone, Warfarin, and Cyclosporine. This suggests that these drugs might have more versatile applications or better bioavailability.

Various indication classes are represented in the dataset, including Glucocorticoid, Immunosuppressant, Anticoagulant, Antihistaminic, Anti-Emetic, Antidepressant, and more. This highlights the diversity of the drugs' applications in treating different medical conditions.

Most of the drugs have chiral structures (chirality = 1), which could impact their pharmacological properties and the way they interact with the biological targets.

Most drugs have no black box warning (black_box_warning = 0), suggesting they have relatively lower risks of severe side effects or complications. However, a few drugs, like Warfarin and Thalidomide, have black box warnings, indicating a higher risk associated with their use.

Overall, the dataset provides insights into the various properties of the drugs, their effectiveness as mono or combination therapies, and their importance within the network. Warfarin stands out as a highly connected and influential drug, while other drugs like Dexamethasone and Tacrolimus also have numerous connections with other drugs. This information can be useful for understanding drug interactions, developing better treatment strategies, and identifying potential new drug combinations.

# Get the subgraphs
subgraphs = nx.connected_components(G)

# Create a figure for the plot
fig = go.Figure()

# Iterate over each subgraph
for i, nodes in enumerate(subgraphs):

    # Convert the set of nodes to a subgraph
    subgraph = G.subgraph(nodes)

    # Get the degree distribution
    degree_dist = nx.degree_histogram(subgraph)

    # Convert the degree distribution to a list of tuples
    degree_dist = [(deg, count) for deg, count in enumerate(degree_dist)]

    # Add a trace for the degree distribution of the current subgraph
    fig.add_trace(go.Scatter(x=[x[0] for x in degree_dist], y=[x[1] for x in degree_dist], mode='lines', name=f'Subgraph {i+1}'))

# Update the layout 
fig.update_layout(xaxis_title='Degree', yaxis_title='Count', title='Degree distribution per subgraph')

# Show the plot
fig.show()

# extentd the dataframe with the degree and betweeness 
# filter by subgraph 1 and sort by degree, show columns degree, Drug_original, Ranking Mono,Ranking Mono,oral,parenteral,indication_class,topical,chirality,availability_type,black_box_warning,first_in_class	ATC_Classification
nodes[nodes['subgraph'] == 1].sort_values('degree', ascending=False)[['degree', 'Drug_original', 'Ranking Mono ','Ranking Combi', 'oral','parenteral', 'indication_class', 'topical', 'chirality', 'availability_type', 'black_box_warning', 'first_in_class', 'ATC1_names','ATC_Classifications']].head(15)

	degree	Drug_original	Ranking Mono	Ranking Combi	oral	parenteral	indication_class	topical	chirality	availability_type	black_box_warning	first_in_class	ATC1_names	ATC_Classifications
0	910	Dexamethasone	75.0	44.0	True	True	Glucocorticoid	True	1.0	1.0	0.0	0.0	{'RESPIRATORY SYSTEM': 2, 'ALIMENTARY TRACT AN...	R01AD53, A01AC02, R01AD03, S02BA06, S03BA01, D...
417	866	Tacrolimus	159.0	248.0	True	True	Immunosuppressant	True	1.0	1.0	1.0	0.0	{}	0
580	864	Warfarin	0.0	0.0	True	True	Anticoagulant	False	0.0	1.0	1.0	0.0	{'BLOOD AND BLOOD FORMING ORGANS': 1}	B01AA03
37	832	Promethazine	0.0	0.0	True	True	Anti-Emetic, Antihistaminic	True	0.0	1.0	1.0	0.0	{'RESPIRATORY SYSTEM': 2, 'DERMATOLOGICALS': 1}	R06AD52, D04AA10, R06AD02
698	822	Clarithromycin	214.0	307.0	True	False	Antibacterial	False	1.0	1.0	0.0	0.0	{'ANTIINFECTIVES FOR SYSTEMIC USE': 1}	J01FA09
484	816	Cyclosporine	0.0	0.0	True	True	Immunosuppressant	True	1.0	1.0	1.0	0.0	{'SENSORY ORGANS': 1, 'ANTINEOPLASTIC AND IMMU...	S01XA18, L04AD01
409	805	Prednisone	85.0	157.0	True	False	Glucocorticoid	False	1.0	1.0	0.0	0.0	{'ALIMENTARY TRACT AND METABOLISM': 1, 'SYSTEM...	A07EA03, H02AB07
239	785	Thalidomide	0.0	0.0	True	False	Sedative-Hypnotic	False	0.0	1.0	1.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}	L04AX02
74	783	Fluconazole	107.0	188.0	True	True	Antifungal	False	2.0	1.0	0.0	0.0	{'ANTIINFECTIVES FOR SYSTEMIC USE': 1, 'DERMAT...	J02AC01, D01AC15
529	781	Methotrexate	31.0	80.0	False	False	Antineoplastic	False	-1.0	-1.0	0.0	-1.0	{}	0
365	754	Erythromycin	340.0	440.0	True	False	Antibacterial	True	1.0	1.0	0.0	0.0	{'SENSORY ORGANS': 1, 'ANTIINFECTIVES FOR SYST...	S01AA17, J01FA01, D10AF02, D10AF52
28	753	Salbutamol	32.0	72.0	True	False	Bronchodilator	True	0.0	1.0	1.0	0.0	{'RESPIRATORY SYSTEM': 2}	R03AC02, R03CC02
20	751	Prednisolone	53.0	58.0	True	False	Glucocorticoid	True	1.0	1.0	0.0	0.0	{'SENSORY ORGANS': 4, 'DERMATOLOGICALS': 2, 'R...	S02BA03, D07XA02, R01AD02, D07AA03, R01AD52, C...
318	734	Doxepin	1271.0	1354.0	True	False	Antidepressant	True	2.0	1.0	1.0	0.0	{'DERMATOLOGICALS': 1, 'NERVOUS SYSTEM': 1}	D04AX01, N06AA12
1508	733	Ozanimod	831.0	895.0	True	False	0	False	1.0	1.0	0.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}	L04AA38

# Filter by subgraph 1 and sort by Top 20 closeness centrality, show columns closeness_centrality, Drug_original, Ranking Mono,Ranking Mono,oral,parenteral,indication_class,topical,chirality,availability_type,black_box_warning,first_in_class	ATC_Classification
nodes[nodes['subgraph'] == 1].sort_values('degree_centrality', ascending=False)[['degree_centrality' ,'Drug_original', 'Ranking Mono ','Ranking Combi', 'oral','parenteral', 'indication_class', 'topical', 'chirality', 'availability_type', 'black_box_warning', 'first_in_class', 'ATC1_names']].head(15)

	degree_centrality	Drug_original	Ranking Mono	Ranking Combi	oral	parenteral	indication_class	topical	chirality	availability_type	black_box_warning	first_in_class	ATC1_names
580	0.025057	Warfarin	0.0	0.0	True	True	Anticoagulant	False	0.0	1.0	1.0	0.0	{'BLOOD AND BLOOD FORMING ORGANS': 1}
1508	0.015335	Ozanimod	831.0	895.0	True	False	0	False	1.0	1.0	0.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}
705	0.012598	Idelalisib	1158.0	1187.0	True	False	0	False	1.0	1.0	1.0	1.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}
239	0.012547	Thalidomide	0.0	0.0	True	False	Sedative-Hypnotic	False	0.0	1.0	1.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}
529	0.012522	Methotrexate	31.0	80.0	False	False	Antineoplastic	False	-1.0	-1.0	0.0	-1.0	{}
484	0.011952	Cyclosporine	0.0	0.0	True	True	Immunosuppressant	True	1.0	1.0	1.0	0.0	{'SENSORY ORGANS': 1, 'ANTINEOPLASTIC AND IMMU...
0	0.011539	Dexamethasone	75.0	44.0	True	True	Glucocorticoid	True	1.0	1.0	0.0	0.0	{'RESPIRATORY SYSTEM': 2, 'ALIMENTARY TRACT AN...
1280	0.010379	Picosulfuric acid	0.0	0.0	False	False	0	False	0.0	0.0	0.0	0.0	{}
657	0.009685	Chloramphenicol	0.0	0.0	True	True	Antibacterial, Antirickettsial	True	1.0	-1.0	0.0	0.0	{'SENSORY ORGANS': 3, 'DERMATOLOGICALS': 2, 'G...
37	0.009564	Promethazine	0.0	0.0	True	True	Anti-Emetic, Antihistaminic	True	0.0	1.0	1.0	0.0	{'RESPIRATORY SYSTEM': 2, 'DERMATOLOGICALS': 1}
318	0.008577	Doxepin	1271.0	1354.0	True	False	Antidepressant	True	2.0	1.0	1.0	0.0	{'DERMATOLOGICALS': 1, 'NERVOUS SYSTEM': 1}
1346	0.008372	Carfilzomib	664.0	743.0	False	True	0	False	1.0	1.0	0.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}
195	0.008225	Morphine	79.0	150.0	True	True	Analgesic (narcotic)	False	1.0	1.0	1.0	0.0	{'NERVOUS SYSTEM': 2, 'ALIMENTARY TRACT AND ME...
417	0.008064	Tacrolimus	159.0	248.0	True	True	Immunosuppressant	True	1.0	1.0	1.0	0.0	{}
28	0.007588	Salbutamol	32.0	72.0	True	False	Bronchodilator	True	0.0	1.0	1.0	0.0	{'RESPIRATORY SYSTEM': 2}

5.5. Eigenvector centrality

# Calculate the closeness centrality for each node
bb = nx.eigenvector_centrality(G)

# Create a histogram of the closeness centrality values
fig = go.Figure(data=[go.Histogram(x=list(bb.values()))])

# Add labels and title to the plot
fig.update_layout(
    title="Eigenvector Centrality Distribution",
    xaxis_title="Eigenvector Centrality",
    yaxis_title="Count"
)

# Show the plot
fig.show()

This dataset shows the eigenvector centrality scores for various drugs, both unweighted and weighted. Eigenvector centrality is a measure of the influence of a node in a network. In this context, it measures the importance of a drug based on its connections to other drugs in the network.

Unweighted eigenvector centrality does not take into account the weights of the connections (edges) between the nodes (drugs). The top 3 drugs based on unweighted eigenvector centrality are Tacrolimus, Dexamethasone, and Clarithromycin. Tacrolimus is an immunosuppressant, Dexamethasone is a glucocorticoid, and Clarithromycin is an antibacterial.

Weighted eigenvector centrality, on the other hand, does take into account the weights of the connections between the nodes. This means that the importance of a drug in the network is affected not only by the number of connections but also by the strength of these connections. The top 3 drugs based on weighted eigenvector centrality are Metronidazole, Dexamethasone, and Prednisone.

Comparing unweighted and weighted eigenvector centrality scores can provide insights into the relative importance of drugs in a network based on different aspects of their connections. In this case, the rankings of the drugs change slightly when the weights of the connections are taken into account.

• Dexamethasone's widespread use in a variety of conditions and its glucocorticoid properties may contribute to its high eigenvector centrality.

• Metronidazole is an antibacterial and antiprotozoal drug used to treat infections caused by anaerobic bacteria and protozoa. Its high weighted eigenvector centrality indicates that it has strong connections with other drugs that also have high centrality scores. The broad spectrum of activity and use of Metronidazole in various infections could be a factor contributing to its high eigenvector centrality, making it a crucial drug to monitor for potential interactions.

• Prednisone is a glucocorticoid drug used to treat various conditions, including allergies, autoimmune diseases, and inflammatory disorders. Its high weighted eigenvector centrality suggests that it has strong connections with other drugs with high centrality scores. This could be attributed to its widespread use and glucocorticoid properties. Similar to Dexamethasone, Prednisone's role in managing a wide range of conditions makes it a key drug in the context of drug interactions.

Conclusion: Drugs with high weighted eigenvector centrality scores, such as Dexamethasone, Metronidazole, and Prednisone, are not only connected to many other drugs but also to other drugs with high centrality scores. This highlights their importance in drug interactions. These drugs are used to treat various conditions, including allergies, infections, autoimmune diseases, and inflammatory disorders. Healthcare providers should be cautious when prescribing these medications in combination with other drugs to minimize the risk of adverse effects and ensure the safety and effectiveness of treatment.

# Filter by subgraph 1 and sort by Top 20 closeness centrality, show columns closeness_centrality, Drug_original, Ranking Mono,Ranking Mono,oral,parenteral,indication_class,topical,chirality,availability_type,black_box_warning,first_in_class	ATC_Classification
nodes[nodes['subgraph'] == 1].sort_values('eigenvector_centrality', ascending=False)[['eigenvector_centrality', 'Drug_original', 'Ranking Mono ','Ranking Combi', 'oral','parenteral', 'indication_class', 'topical', 'chirality', 'availability_type', 'black_box_warning', 'first_in_class', 'ATC1_names']].head(10)

	eigenvector_centrality	Drug_original	Ranking Mono	Ranking Combi	oral	parenteral	indication_class	topical	chirality	availability_type	black_box_warning	first_in_class	ATC1_names
417	0.078592	Tacrolimus	159.0	248.0	True	True	Immunosuppressant	True	1.0	1.0	1.0	0.0	{}
0	0.077820	Dexamethasone	75.0	44.0	True	True	Glucocorticoid	True	1.0	1.0	0.0	0.0	{'RESPIRATORY SYSTEM': 2, 'ALIMENTARY TRACT AN...
698	0.076040	Clarithromycin	214.0	307.0	True	False	Antibacterial	False	1.0	1.0	0.0	0.0	{'ANTIINFECTIVES FOR SYSTEMIC USE': 1}
74	0.074010	Fluconazole	107.0	188.0	True	True	Antifungal	False	2.0	1.0	0.0	0.0	{'ANTIINFECTIVES FOR SYSTEMIC USE': 1, 'DERMAT...
37	0.072799	Promethazine	0.0	0.0	True	True	Anti-Emetic, Antihistaminic	True	0.0	1.0	1.0	0.0	{'RESPIRATORY SYSTEM': 2, 'DERMATOLOGICALS': 1}
365	0.071019	Erythromycin	340.0	440.0	True	False	Antibacterial	True	1.0	1.0	0.0	0.0	{'SENSORY ORGANS': 1, 'ANTIINFECTIVES FOR SYST...
484	0.070463	Cyclosporine	0.0	0.0	True	True	Immunosuppressant	True	1.0	1.0	1.0	0.0	{'SENSORY ORGANS': 1, 'ANTINEOPLASTIC AND IMMU...
409	0.070026	Prednisone	85.0	157.0	True	False	Glucocorticoid	False	1.0	1.0	0.0	0.0	{'ALIMENTARY TRACT AND METABOLISM': 1, 'SYSTEM...
28	0.069195	Salbutamol	32.0	72.0	True	False	Bronchodilator	True	0.0	1.0	1.0	0.0	{'RESPIRATORY SYSTEM': 2}
239	0.069081	Thalidomide	0.0	0.0	True	False	Sedative-Hypnotic	False	0.0	1.0	1.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}

# add neighbors in the nodes dataframe
nodes['neighbors'] = nodes['DDInterID'].apply(lambda x: list(G.neighbors(x)))

# Filter by subgraph 1 and sort by Top 20 closeness centrality, show columns closeness_centrality, Drug_original, Ranking Mono,Ranking Mono,oral,parenteral,indication_class,topical,chirality,availability_type,black_box_warning,first_in_class	ATC_Classification
nodes[nodes['subgraph'] == 1].sort_values('weighted_eigenvector_centrality', ascending=False)[['weighted_eigenvector_centrality', 'Drug_original', 'Ranking Mono ','Ranking Combi', 'oral','parenteral', 'indication_class', 'topical', 'chirality', 'availability_type', 'black_box_warning', 'first_in_class', 'ATC1_names']].head(10)

	weighted_eigenvector_centrality	Drug_original	Ranking Mono	Ranking Combi	oral	parenteral	indication_class	topical	chirality	availability_type	black_box_warning	first_in_class	ATC1_names
0	0.081356	Dexamethasone	75.0	44.0	True	True	Glucocorticoid	True	1.0	1.0	0.0	0.0	{'RESPIRATORY SYSTEM': 2, 'ALIMENTARY TRACT AN...
626	0.076586	Metronidazole	152.0	234.0	True	True	Antibacterial,Antiprotozoal (trichomonas)	True	2.0	1.0	1.0	0.0	{'DERMATOLOGICALS': 1, 'ANTIPARASITIC PRODUCTS...
409	0.076049	Prednisone	85.0	157.0	True	False	Glucocorticoid	False	1.0	1.0	0.0	0.0	{'ALIMENTARY TRACT AND METABOLISM': 1, 'SYSTEM...
37	0.075555	Promethazine	0.0	0.0	True	True	Anti-Emetic, Antihistaminic	True	0.0	1.0	1.0	0.0	{'RESPIRATORY SYSTEM': 2, 'DERMATOLOGICALS': 1}
28	0.075117	Salbutamol	32.0	72.0	True	False	Bronchodilator	True	0.0	1.0	1.0	0.0	{'RESPIRATORY SYSTEM': 2}
417	0.073848	Tacrolimus	159.0	248.0	True	True	Immunosuppressant	True	1.0	1.0	1.0	0.0	{}
74	0.073254	Fluconazole	107.0	188.0	True	True	Antifungal	False	2.0	1.0	0.0	0.0	{'ANTIINFECTIVES FOR SYSTEMIC USE': 1, 'DERMAT...
698	0.072693	Clarithromycin	214.0	307.0	True	False	Antibacterial	False	1.0	1.0	0.0	0.0	{'ANTIINFECTIVES FOR SYSTEMIC USE': 1}
20	0.072523	Prednisolone	53.0	58.0	True	False	Glucocorticoid	True	1.0	1.0	0.0	0.0	{'SENSORY ORGANS': 4, 'DERMATOLOGICALS': 2, 'R...
580	0.071685	Warfarin	0.0	0.0	True	True	Anticoagulant	False	0.0	1.0	1.0	0.0	{'BLOOD AND BLOOD FORMING ORGANS': 1}

5.6.Clustering Coefficient

# Calculate the closeness centrality for each node
cc = nx.clustering(G)

# Create a histogram of the closeness centrality values
fig = go.Figure(data=[go.Histogram(x=list(cc.values()))])

# Add labels and title to the plot
fig.update_layout(
    title="Clustering Coefficient Distribution",
    xaxis_title="Clustering Coefficient",
    yaxis_title="Count"
)

# Show the plot
fig.show()

Drugs with a clustering coefficient of 1.000000 seem to have no significant interactions or low severity interactions with other drugs, which might indicate that these drugs are safe to use with other medications. Among these drugs, Tropicamide is an anticholinergic used in sensory organs, Azelaic acid is an anti-acne medication, Inulin is a diagnostic aid for renal function determination, Acetohydroxamic acid is an enzyme inhibitor, and Pegademase is an antineoplastic and immunomodulating agent.

Simethicone is taken off from the market and you can see that most of these Drugs are not represented in the market with the ranking mono or combi as zero.

The clustering coefficient can give us an idea of how "connected" a drug is within the network of drug interactions. A higher clustering coefficient indicates that the drug is more likely to interact with other drugs. In the case of weighted clustering coefficient, a higher value suggests that the drug has more diverse interactions of varying severity levels. It is important to consider these coefficients when prescribing medications to ensure patient safety and minimize adverse drug interactions.

# Filter by subgraph 1 and sort by Top 2x0 closeness centrality, show columns closeness_centrality, Drug_original, Ranking Mono,Ranking Mono,oral,parenteral,indication_class,topical,chirality,availability_type,black_box_warning,first_in_class	ATC_Classification
nodes[nodes['subgraph'] == 1].sort_values('clustering_coefficient', ascending=False)[['clustering_coefficient', 'Drug_original', 'Ranking Mono ','Ranking Combi', 'oral','parenteral', 'indication_class', 'topical', 'chirality', 'availability_type', 'black_box_warning', 'first_in_class', 'ATC1_names']].head(10)

	clustering_coefficient	Drug_original	Ranking Mono	Ranking Combi	oral	parenteral	indication_class	topical	chirality	availability_type	black_box_warning	first_in_class	ATC1_names
1155	1.000000	Tropicamide	431.0	518.0	False	False	Anticholinergic	True	0.0	1.0	0.0	0.0	{'SENSORY ORGANS': 2}
1169	1.000000	Azelaic acid	0.0	0.0	False	False	Anti-Acne	True	2.0	1.0	0.0	0.0	{'DERMATOLOGICALS': 1}
1464	1.000000	Inulin	0.0	0.0	False	True	Diagnostic Aid (renal function determination)	False	1.0	0.0	0.0	0.0	{}
1674	1.000000	Acetohydroxamic acid	0.0	0.0	True	False	Enzyme Inhibitor (urease)	False	2.0	1.0	0.0	0.0	{'GENITO URINARY SYSTEM AND SEX HORMONES': 1}
1656	1.000000	Pegademase	0.0	0.0	False	False	0	False	1.0	-1.0	0.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}
1927	1.000000	Simethicone	0.0	0.0	True	False	0	False	2.0	2.0	0.0	0.0	{}
1477	1.000000	Secretin human	0.0	0.0	False	True	0	False	1.0	1.0	0.0	0.0	{}
1681	1.000000	Creatine	0.0	0.0	False	False	0	False	2.0	-1.0	0.0	0.0	{}
1196	0.996377	Levobunolol	1032.0	1077.0	False	False	Anti-Adrenergic (beta-receptor)	True	1.0	1.0	0.0	0.0	{'SENSORY ORGANS': 1}
1840	0.995402	gamma-Aminobutyric acid	0.0	0.0	False	False	0	False	2.0	-1.0	0.0	0.0	{'NERVOUS SYSTEM': 1}

Focus on the market which has been used only. Filter by Ranking combi/mono >0

• Tropicamide: An anticholinergic drug with a high clustering coefficient of 1.000000, indicating low or no significant interactions with other drugs. It is primarily used in the treatment of eye disorders and is classified under 'SENSORY ORGANS'. Tropicamide is a topical medication, and its chirality is 0.0 (racemic mixture).

• Levobunolol: This drug is an anti-adrenergic (beta-receptor) medication, also used in sensory organs, with a clustering coefficient of 0.996377. It is a topical medication with a chirality of 1.0 (single stereoisomer).

• Travoprost: A medication used for sensory organs with a clustering coefficient of 0.992347. It is a topical medication with a chirality of 1.0 (single stereoisomer).

• Brinzolamide: An antiglaucoma agent used in sensory organs with a clustering coefficient of 0.992063. It is a topical medication with a chirality of 1.0 (single stereoisomer).

• Nitrazepam: A sedative-hypnotic and anticonvulsant drug with a clustering coefficient of 0.959596. It belongs to the 'NERVOUS SYSTEM' category and is an oral medication with a chirality of 2.0 (archiral molecule).

The interpretations for these drugs are as follows:

Most of these drugs are used for sensory organs or the nervous system, and they have high clustering coefficients, suggesting low or no significant interactions with other drugs. The majority of these drugs are either topical or oral medications. Chirality varies across these drugs, with some being racemic mixtures, single stereoisomers, or archiral molecules. Chirality may influence the drug's pharmacokinetics, efficacy, and safety profile. In conclusion, these drugs are generally safe to use with other medications due to their high clustering coefficients. However, it is always essential to consider the individual patient's medical history, potential drug interactions, and the drug's chirality when prescribing medications.

# Filter by subgraph 1  and ranking mono > 0 and sort by Top 20 closeness centrality, show columns closeness_centrality, Drug_original, Ranking Mono,Ranking Mono,oral,parenteral,indication_class,topical,chirality,availability_type,black_box_warning,first_in_class	ATC_Classification
nodes[(nodes['subgraph'] == 1) & (nodes['Ranking Mono '] > 0)].sort_values('clustering_coefficient', ascending=False)[['clustering_coefficient', 'Drug_original', 'Ranking Mono ','Ranking Combi', 'oral','parenteral', 'indication_class', 'topical', 'chirality', 'availability_type', 'black_box_warning', 'first_in_class', 'ATC1_names']].head(10)

	clustering_coefficient	Drug_original	Ranking Mono	Ranking Combi	oral	parenteral	indication_class	topical	chirality	availability_type	black_box_warning	first_in_class	ATC1_names
1155	1.000000	Tropicamide	431.0	518.0	False	False	Anticholinergic	True	0.0	1.0	0.0	0.0	{'SENSORY ORGANS': 2}
1196	0.996377	Levobunolol	1032.0	1077.0	False	False	Anti-Adrenergic (beta-receptor)	True	1.0	1.0	0.0	0.0	{'SENSORY ORGANS': 1}
1123	0.992347	Travoprost	474.0	427.0	False	False	0	True	1.0	1.0	0.0	0.0	{'SENSORY ORGANS': 1}
1201	0.992063	Brinzolamide	330.0	319.0	False	False	Antiglaucoma Agent	True	1.0	1.0	0.0	0.0	{'SENSORY ORGANS': 2}
1209	0.959596	Nitrazepam	714.0	536.0	False	False	Sedative-Hypnotic, Anticonvulsant	False	2.0	-1.0	0.0	0.0	{'NERVOUS SYSTEM': 1}
1160	0.928365	Bimatoprost	380.0	354.0	False	False	0	True	1.0	1.0	0.0	0.0	{'SENSORY ORGANS': 1}
1178	0.923810	Acamprosate	836.0	900.0	True	False	0	False	2.0	1.0	0.0	0.0	{'NERVOUS SYSTEM': 1}
1124	0.920378	Etomidate	996.0	1047.0	False	True	Sedative-Hypnotic	False	1.0	1.0	0.0	0.0	{'NERVOUS SYSTEM': 1}
1121	0.913753	Oseltamivir	735.0	809.0	False	False	0	False	1.0	-1.0	0.0	0.0	{}
1669	0.908135	Isatuximab	981.0	1029.0	False	True	0	False	1.0	1.0	0.0	0.0	{'ANTINEOPLASTIC AND IMMUNOMODULATING AGENTS': 1}

6. Drug Graph Metrics

6.1.Transitivity

This algorithm calculates the proportion of all possible triangles in the graph that are closed. It can help identify the prevalence of clustering or how connected the nodes are within the network.

The value 0.3258070626776695 is the transitivity or global clustering coefficient of the graph G.

Transitivity measures the fraction of "triangles" or three-node subgraphs that are completed in a network. In other words, it measures how likely two nodes that are connected to a common neighbor are also connected to each other. The higher the transitivity value, the more likely it is for two nodes to be connected if they share a neighbor.

In the case of your graph G, the transitivity value of 0.3258 indicates that there is some clustering or grouping of nodes, but the graph is not highly clustered. This means that while there are some connections between nodes that form triangles, there are also many nodes that do not form triangles, and the graph has a relatively low level of overall clustering.

triadic_closure = nx.transitivity(G)
triadic_closure

0.3258070626776695

6.2. K-Core decomposition

K-core decomposition is a network analysis technique used to identify densely connected subgraphs in a graph. It is different from other centrality measures like degree and closeness centrality in that it identifies groups of nodes that are highly connected to each other, rather than focusing on individual nodes' connectivity. K-core decomposition can reveal the structure of a network and help identify highly connected communities or clusters within the network.

Drugs with higher core numbers (e.g., Dexamethasone, Ivacaftor, Mannitol, Betamethasone, and Budesonide) are likely to be more central to the structure of the network and have more connections to other drugs. This could mean that these drugs are involved in more drug interactions or have more significant effects on the network as a whole. These drugs have many connections, indicating that they may interact with a wide range of other drugs or have multiple therapeutic applications. This information could be useful in drug repositioning efforts or understanding potential drug-drug interactions.

On the other hand, drugs with lower core numbers (e.g., Manganese (sulfate) and Urofollitropin) may be less important or less central to the network.The nodes with the lowest core numbers have a core number of 1, indicating that these drugs are only connected to one other drug in the network. These drugs may have more specific therapeutic applications or fewer known interactions with other drugs

G = nx.from_pandas_edgelist(edges, source='Drug_A', target='Drug_B', edge_attr='Weight', create_using=nx.Graph())
k_cores = nx.core_number(G)

# Sort the dictionary by values in descending order
sorted_k_cores = sorted(k_cores.items(), key=operator.itemgetter(1), reverse=True)

# Get the top 10 nodes with the highest core numbers
top_10_highest = sorted_k_cores[:10]
print("Nodes with the highest core numbers:")
for node, core_num in top_10_highest:
    print(f"{node}: {core_num}")

# Get the top 10 nodes with the lowest core numbers
top_10_lowest = sorted_k_cores[-10:]
print("Nodes with the lowest core numbers:")
for node, core_num in top_10_lowest:
    print(f"{node}: {core_num}")

Nodes with the highest core numbers:
Dexamethasone: 150
Ivacaftor: 150
Mannitol: 150
Betamethasone: 150
Budesonide: 150
Flurbiprofen: 150
Hydrocortisone: 150
Ibuprofen: 150
Mercaptopurine: 150
Prednisolone: 150
Nodes with the lowest core numbers:
Manganese (sulfate): 1
Manganese (chloride): 1
Triheptanoin: 1
Chorionic Gonadotropin (Human): 1
Menotropins: 1
Follitropin: 1
Urofollitropin: 1
Sodium glycerophosphate: 1
Citrus bioflavonoids: 1
Pyrophosphoric acid: 1

import networkx as nx
import matplotlib.pyplot as plt

def draw_k_cores(G, k_cores, k):
    k_core_subgraph = nx.k_core(G, k=k)
    node_colors = [k_cores[node] for node in k_core_subgraph.nodes]

    plt.figure(figsize=(12, 8))
    plt.title(f"{k}-core subgraph")
    pos = nx.spring_layout(k_core_subgraph)
    nx.draw(k_core_subgraph, pos, node_size=300, node_color=node_colors, cmap=plt.cm.viridis, font_size=4, font_weight='bold')
    plt.show()

G = nx.from_pandas_edgelist(edges, source='Drug_A', target='Drug_B', edge_attr='Weight', create_using=nx.Graph())
k_cores = nx.core_number(G)

# Visualize k-core decomposition for specific k values
k_values = [1, 3, 5,20]
for k in k_values:
    draw_k_cores(G, k_cores, k)

Drugs with high core numbers and high edge weights, such as Dexamethasone, Ivacaftor, Mannitol, Betamethasone, Budesonide, Flurbiprofen, Hydrocortisone, Ibuprofen, Mercaptopurine, and Prednisolone, are likely to be the most influential and potentially the most harmful drugs in the network. They have a large number of connections to other drugs, and their interactions are likely to be severe. Drugs with low core numbers and high edge weights, such as Manganese (sulfate), Manganese (chloride), Triheptanoin, Chorionic Gonadotropin (Human), Menotropins, Follitropin, Urofollitropin, Sodium glycerophosphate, Citrus bioflavonoids, and Pyrophosphoric acid, may not be as central in the network, but their interactions with other drugs could alsp be likely to be severe or not known. Drugs with high core numbers and low edge weights, if any, may have a large number of connections to other drugs, but their effects on the network may not be as significant. It's possible that some drugs in the network may fall into this category. Drugs with low core numbers and low edge weights, such as other drugs in the network that are not mentioned in the top or bottom 10, are likely to be less important and less harmful in the network. They have fewer connections to other drugs, and their interactions may not be particularly severe

6.3.Dispersion

This algorithm computes the dispersion between two nodes, which can help identify pairs of nodes that are not directly connected but have many common neighbors.

import itertools
import networkx as nx

def dispersion(G, u, v):
    shortest_path_length = nx.shortest_path_length(G, u, v)
    neighbor_count = len(set(G.neighbors(u)) | set(G.neighbors(v)))
    return shortest_path_length / neighbor_count

dispersion_dict = {}
for u, v in itertools.combinations(G.nodes(), 2):
    # check if there is a path between u and v
    if not nx.has_path(G, u, v):
        continue
    dispersion_value = dispersion(G, u, v)
    if dispersion_value > 0:
        key = (u, v) if u < v else (v, u)
        dispersion_dict[key] = dispersion_value

top_10_pairs = sorted(dispersion_dict.items(), key=lambda x: x[1], reverse=True)[:10]

print("Top 10 pairs with highest dispersion values:")
for i, (pair, value) in enumerate(top_10_pairs):
    print(f"{i+1}. Pair: {pair}, Dispersion value: {value}")

Top 10 pairs with highest dispersion values:
1. Pair: ('Lutein', 'Silver (topical)'), Dispersion value: 3.5
2. Pair: ('Hexaminolevulinate', 'Silver (topical)'), Dispersion value: 3.5
3. Pair: ('Norelgestromin', 'Silver (topical)'), Dispersion value: 3.5
4. Pair: ('Gonadorelin', 'Silver (topical)'), Dispersion value: 3.5
5. Pair: ('Pegaptanib (ophthalmic)', 'Silver (topical)'), Dispersion value: 3.5
6. Pair: ('Chorionic Gonadotropin (Human)', 'Silver (topical)'), Dispersion value: 3.5
7. Pair: ('Menotropins', 'Silver (topical)'), Dispersion value: 3.5
8. Pair: ('Follitropin', 'Silver (topical)'), Dispersion value: 3.5
9. Pair: ('Silver (topical)', 'Urofollitropin'), Dispersion value: 3.5
10. Pair: ('Lutein', 'Thimerosal (topical)'), Dispersion value: 3.5

The dispersion output indicates the strength of the relationships between pairs of drugs. A high dispersion value implies that the drugs in the pair are not only connected but also share connections to other drugs within the network. In other words, high dispersion values suggest that these pairs of drugs are involved in a wider network of drug interactions.

In this case, it seems that Silver (topical) is a common factor in the top 9 pairs with the highest dispersion values. Silver (topical) is often used for its antimicrobial properties in wound care and skin infections. This suggests that Silver (topical) may have a higher likelihood of interacting with a variety of drugs, as indicated by the high dispersion values.

Additionally, several drugs in the pairs, such as Lutein, Gonadorelin, Menotropins, and Follitropin, are related to hormones or hormone regulation. This may indicate that hormonal drugs or drugs that affect hormone regulation might have a higher propensity to interact with Silver (topical).

The 10th pair, ('Lutein', 'Thimerosal (topical)'), also has a high dispersion value. Thimerosal is a mercury-containing organic compound used as a preservative in some vaccines and other medical products. This suggests that Lutein, a carotenoid found in various dietary supplements, might have a higher likelihood of interacting with Thimerosal (topical).

In conclusion, the dispersion output highlights the potential for drug interactions between Silver (topical) and a variety of other drugs, particularly those related to hormones or hormone regulation. It also suggests that Lutein may have a higher propensity to interact with Thimerosal (topical). These results can be useful for healthcare professionals and researchers in identifying possible drug interactions and guiding decision-making when prescribing or studying these drugs.

if G.has_edge('Lutein', 'Silver (topical)'):
    print(" Both are  neighbors")
else:
    print("Both are not neighbors")

Both are not neighbors

6.4. Assortativity coefficient

This measure computes the tendency of nodes with similar attributes (e.g., drug properties) to be connected. A positive assortativity coefficient indicates that nodes with similar properties tend to be connected, while a negative value indicates the opposite.

nodes.head(2)

	DDInterID	Drug	Drug_original	Units MAT Feb 2023	% Growth	Sum incl combi	Ranking Mono	Ranking Combi	topical	nasal	ophthalmic	oral	parenteral	indication_class	chirality	black_box_warning	availability_type	first_in_class	ATC_Classifications	ATC1_names	ANTIINFECTIVES_FOR_SYSTEMIC_USE	RESPIRATORY_SYSTEM	ANTIPARASITIC_PRODUCTS,_INSECTICIDES_AND_REPELLENTS	ANTINEOPLASTIC_AND_IMMUNOMODULATING_AGENTS	MUSCULO-SKELETAL_SYSTEM	ALIMENTARY_TRACT_AND_METABOLISM	BLOOD_AND_BLOOD_FORMING_ORGANS	SENSORY_ORGANS	SYSTEMIC_HORMONAL_PREPARATIONS,_EXCL._SEX_HORMONES_AND_INSULINS	NERVOUS_SYSTEM	GENITO_URINARY_SYSTEM_AND_SEX_HORMONES	DERMATOLOGICALS	VARIOUS	CARDIOVASCULAR_SYSTEM	ATC2_names	R01_-_NASAL_PREPARATIONS	A01_-_STOMATOLOGICAL_PREPARATIONS	S02_-_OTOLOGICALS	S03_-_OPHTHALMOLOGICAL_AND_OTOLOGICAL_PREPARATIONS	D07_-_CORTICOSTEROIDS,_DERMATOLOGICAL_PREPARATIONS	D10_-_ANTI-ACNE_PREPARATIONS	S01_-_OPHTHALMOLOGICALS	C05_-_VASOPROTECTIVES	H02_-_CORTICOSTEROIDS_FOR_SYSTEMIC_USE	R07_-_OTHER_RESPIRATORY_SYSTEM_PRODUCTS	J05_-_ANTIVIRALS_FOR_SYSTEMIC_USE	R03_-_DRUGS_FOR_OBSTRUCTIVE_AIRWAY_DISEASES	A07_-_ANTIDIARRHEALS,_INTESTINAL_ANTIINFLAMMATORY/ANTIINFECTIVE_AGENTS	R06_-_ANTIHISTAMINES_FOR_SYSTEMIC_USE	N06_-_PSYCHOANALEPTICS	D11_-_OTHER_DERMATOLOGICAL_PREPARATIONS	N02_-_ANALGESICS	R05_-_COUGH_AND_COLD_PREPARATIONS	N07_-_OTHER_NERVOUS_SYSTEM_DRUGS	P03_-_ECTOPARASITICIDES,_INCL._SCABICIDES,_INSECTICIDES_AND_REPELLENTS	R02_-_THROAT_PREPARATIONS	D04_-_ANTIPRURITICS,_INCL._ANTIHISTAMINES,_ANESTHETICS,_ETC.	N01_-_ANESTHETICS	C01_-_CARDIAC_THERAPY	C04_-_PERIPHERAL_VASODILATORS	L04_-_IMMUNOSUPPRESSANTS	B01_-_ANTITHROMBOTIC_AGENTS	L02_-_ENDOCRINE_THERAPY	L01_-_ANTINEOPLASTIC_AGENTS	A11_-_VITAMINS	M02_-_TOPICAL_PRODUCTS_FOR_JOINT_AND_MUSCULAR_PAIN	M01_-_ANTIINFLAMMATORY_AND_ANTIRHEUMATIC_PRODUCTS	G02_-_OTHER_GYNECOLOGICALS	A10_-_DRUGS_USED_IN_DIABETES	C07_-_BETA_BLOCKING_AGENTS	J02_-_ANTIMYCOTICS_FOR_SYSTEMIC_USE	D01_-_ANTIFUNGALS_FOR_DERMATOLOGICAL_USE	N05_-_PSYCHOLEPTICS	H01_-_PITUITARY_AND_HYPOTHALAMIC_HORMONES_AND_ANALOGUES	N03_-_ANTIEPILEPTICS	M03_-_MUSCLE_RELAXANTS	N04_-_ANTI-PARKINSON_DRUGS	G04_-_UROLOGICALS	A03_-_DRUGS_FOR_FUNCTIONAL_GASTROINTESTINAL_DISORDERS	C02_-_ANTIHYPERTENSIVES	A04_-_ANTIEMETICS_AND_ANTINAUSEANTS	D08_-_ANTISEPTICS_AND_DISINFECTANTS	V03_-_ALL_OTHER_THERAPEUTIC_PRODUCTS	A08_-_ANTIOBESITY_PREPARATIONS,_EXCL._DIET_PRODUCTS	A12_-_MINERAL_SUPPLEMENTS	B05_-_BLOOD_SUBSTITUTES_AND_PERFUSION_SOLUTIONS	B02_-_ANTIHEMORRHAGICS	L03_-_IMMUNOSTIMULANTS	C03_-_DIURETICS	C09_-_AGENTS_ACTING_ON_THE_RENIN-ANGIOTENSIN_SYSTEM	A02_-_DRUGS_FOR_ACID_RELATED_DISORDERS	J01_-_ANTIBACTERIALS_FOR_SYSTEMIC_USE	D06_-_ANTIBIOTICS_AND_CHEMOTHERAPEUTICS_FOR_DERMATOLOGICAL_USE	A06_-_DRUGS_FOR_CONSTIPATION	P01_-_ANTIPROTOZOALS	J04_-_ANTIMYCOBACTERIALS	V04_-_DIAGNOSTIC_AGENTS	G03_-_SEX_HORMONES_AND_MODULATORS_OF_THE_GENITAL_SYSTEM	M09_-_OTHER_DRUGS_FOR_DISORDERS_OF_THE_MUSCULO-SKELETAL_SYSTEM	G01_-_GYNECOLOGICAL_ANTIINFECTIVES_AND_ANTISEPTICS	C08_-_CALCIUM_CHANNEL_BLOCKERS	M05_-_DRUGS_FOR_TREATMENT_OF_BONE_DISEASES	V08_-_CONTRAST_MEDIA	C10_-_LIPID_MODIFYING_AGENTS	V10_-_THERAPEUTIC_RADIOPHARMACEUTICALS	D05_-_ANTIPSORIATICS	A16_-_OTHER_ALIMENTARY_TRACT_AND_METABOLISM_PRODUCTS	M04_-_ANTIGOUT_PREPARATIONS	B06_-_OTHER_HEMATOLOGICAL_AGENTS	A14_-_ANABOLIC_AGENTS_FOR_SYSTEMIC_USE	P02_-_ANTHELMINTICS	H05_-_CALCIUM_HOMEOSTASIS	A05_-_BILE_AND_LIVER_THERAPY	V09_-_DIAGNOSTIC_RADIOPHARMACEUTICALS	H03_-_THYROID_THERAPY	B03_-_ANTIANEMIC_PREPARATIONS	D09_-_MEDICATED_DRESSINGS	D03_-_PREPARATIONS_FOR_TREATMENT_OF_WOUNDS_AND_ULCERS	D02_-_EMOLLIENTS_AND_PROTECTIVES	J07_-_VACCINES	A09_-_DIGESTIVES,_INCL._ENZYMES	V06_-_GENERAL_NUTRIENTS	H04_-_PANCREATIC_HORMONES	subgraph	closeness_centrality	weighted_closeness_centrality	betweenness_centrality	weighted_betweenness_centrality	pagerank	weighted_pagerank	degree_centrality	eigenvector_centrality	weighted_eigenvector_centrality	clustering_coefficient	weighted_clustering_coefficient	degree	neighbors
0	DDInter513	Dexamethasone	Dexamethasone	401082.0	0.055232	1162374.0	75.0	44.0	True	False	False	True	True	Glucocorticoid	1.0	0.0	1.0	0.0	R01AD53, A01AC02, R01AD03, S02BA06, S03BA01, D...	{'RESPIRATORY SYSTEM': 2, 'ALIMENTARY TRACT AN...	0	2	0	0	0	1	0	4	1	0	0	3	0	1	{'R01 - NASAL PREPARATIONS': 1, 'A01 - STOMATO...	1	1	1	1	1	1	1	1	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.635260	0.304632	0.011539	0.011286	0.002508	0.002468	0.011539	0.077820	0.075080	0.205962	0.148061	910	[DDInter582, DDInter1961, DDInter3, DDInter7, ...
1	DDInter997	Ivacaftor	Ivacaftor	0.0	0.000000	0.0	0.0	0.0	False	False	False	True	0	0	2.0	0.0	1.0	1.0	R07AX02	{'RESPIRATORY SYSTEM': 1}	0	1	0	0	0	0	0	0	0	0	0	0	0	0	{'R07 - OTHER RESPIRATORY SYSTEM PRODUCTS': 1}	0	0	0	0	0	0	0	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.527212	0.255197	0.000709	0.000842	0.001023	0.001022	0.000709	0.037246	0.038606	0.351685	0.262743	377	[DDInter582, DDInter3, DDInter8, DDInter991, D...

#show isnul values
nodes.isnull().sum()

DDInterID                             0
Drug                                  0
Drug_original                         0
Units MAT Feb 2023                    0
 % Growth                             0
                                   ... 
weighted_eigenvector_centrality       0
clustering_coefficient                0
weighted_clustering_coefficient       0
degree                             1939
neighbors                             0
Length: 137, dtype: int64

# Iterate through each column in the nodes dataframe starting from column 8 and calculate the assortativity for that attribute
for column in nodes.columns[9:-15]:
    nx.set_node_attributes(G, nodes.set_index('DDInterID')[column].to_dict(), column)
    assortativity = nx.attribute_assortativity_coefficient(G, column)
    if assortativity > 0.05 or assortativity < -0.05:
        print(f"Assortativity for attribute {column}: {assortativity}")

Assortativity for attribute oral: 0.0810650117431923
Assortativity for attribute ANTINEOPLASTIC_AND_IMMUNOMODULATING_AGENTS: 0.1164132632967772
Assortativity for attribute BLOOD_AND_BLOOD_FORMING_ORGANS: 0.07157906119795782
Assortativity for attribute NERVOUS_SYSTEM: -0.05162355917451893
Assortativity for attribute N01_-_ANESTHETICS: 0.05131632904400924
Assortativity for attribute L04_-_IMMUNOSUPPRESSANTS: 0.08708673115170827
Assortativity for attribute B01_-_ANTITHROMBOTIC_AGENTS: 0.11689660606015763

• Oral (0.08119514025667399): There is a weak tendency for nodes with similar values for the "oral" attribute to connect to each other more often than expected by chance.
• Blood and Blood Forming Organs (0.07157906119795782): There is a weak tendency for nodes with similar values for the "Blood and Blood Forming Organs" attribute to connect to each other more often than expected by chance.
• Nervous System (-0.05162355917451893): There is a weak tendency for nodes with dissimilar values for the "Nervous System" attribute to connect to each other more often than expected by chance.
• Antineoplastic and Immunomodulating Agents (0.1164132632967772): There is a moderate tendency for nodes with similar values for the "Antineoplastic and Immunomodulating Agents" attribute to connect to each other more often than expected by chance.
• N01 - Anesthetics (0.05131632904400924): There is a weak tendency for nodes with similar values for the "N01 - Anesthetics" attribute to connect to each other more often than expected by chance.
• L04 - Immunosuppressants (0.08708673115170827): There is a weak tendency for nodes with similar values for the "L04 - Immunosuppressants" attribute to connect to each other more often than expected by chance.
• B01 - Antithrombotic Agents (0.11689660606015763): There is a moderate tendency for nodes with similar values for the "B01 - Antithrombotic Agents" attribute to connect to each other more often than expected by chance.

In general, positive assortativity indicates that nodes with similar attribute values tend to connect to each other, while negative assortativity indicates that nodes with dissimilar attribute values tend to connect to each other. However, the strength of the assortativity coefficient can range from very weak (close to zero) to very strong (close to 1 or -1). Therefore, you can interpret these results as weak to moderate assortative behavior for the attributes that have assortativity coefficients that are significantly different from zero.

6.5.Jaccard similarity coefficient

This coefficient measures the similarity between sets of neighbors of two nodes, which can be used to predict potential drug interactions.

The Jaccard similarity coefficient measures the similarity between two sets. In the context of a network, it is commonly used to measure the similarity between sets of neighbors of two nodes. The Jaccard similarity coefficient is defined as the size of the intersection of the two sets divided by the size of their union. In other words, it takes into account the number of common neighbors between two nodes and normalizes it by the total number of unique neighbors the nodes have

G = nx.from_pandas_edgelist(edges, source='Drug_A', target='Drug_B', edge_attr='Weight', create_using=nx.Graph())

def jaccard_similarity(u, v, G):
    u_neighbors = set(G.neighbors(u))
    v_neighbors = set(G.neighbors(v))
    intersection = u_neighbors.intersection(v_neighbors)
    union = u_neighbors.union(v_neighbors)
    return len(intersection) / len(union)

import itertools

# Define a function to calculate Jaccard similarity between all pairs of nodes in the graph
def all_jaccard_similarities(G):
    node_pairs = itertools.combinations(G.nodes(), 2)
    similarities = {}
    for u, v in node_pairs:
        similarities[(u, v)] = jaccard_similarity(u, v, G)
    return similarities

# Calculate Jaccard similarities for all pairs of nodes in the graph
similarities = all_jaccard_similarities(G)

# Sort the similarities in descending order
sorted_similarities = sorted(similarities.items(), key=lambda x: x[1], reverse=True)

# Print the top 20 pairs of nodes with the highest Jaccard similarity
for i, (pair, similarity) in enumerate(sorted_similarities[:20]):
    print(f"{i+1}. Nodes with highest Jaccard similarity: {pair}")
    print(f"   Jaccard similarity value: {similarity}")

1. Nodes with highest Jaccard similarity: ('Flurbiprofen (ophthalmic)', 'Diclofenac (ophthalmic)')
   Jaccard similarity value: 1.0
2. Nodes with highest Jaccard similarity: ('Botulinum toxin type A', 'Botulinum Toxin Type B')
   Jaccard similarity value: 1.0
3. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Hydrocortisone (topical)')
   Jaccard similarity value: 1.0
4. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Triamcinolone (topical)')
   Jaccard similarity value: 1.0
5. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Alclometasone (topical)')
   Jaccard similarity value: 1.0
6. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Amcinonide')
   Jaccard similarity value: 1.0
7. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Clobetasol (topical)')
   Jaccard similarity value: 1.0
8. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Clocortolone (topical)')
   Jaccard similarity value: 1.0
9. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Desonide (topical)')
   Jaccard similarity value: 1.0
10. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Desoximetasone (topical)')
   Jaccard similarity value: 1.0
11. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Diflorasone (topical)')
   Jaccard similarity value: 1.0
12. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Fluocinonide (topical)')
   Jaccard similarity value: 1.0
13. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Flurandrenolide (topical)')
   Jaccard similarity value: 1.0
14. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Halcinonide (topical)')
   Jaccard similarity value: 1.0
15. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Ulobetasol (topical)')
   Jaccard similarity value: 1.0
16. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Methylprednisolone (topical)')
   Jaccard similarity value: 1.0
17. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Prednicarbate (topical)')
   Jaccard similarity value: 1.0
18. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Chromic chloride')
   Jaccard similarity value: 1.0
19. Nodes with highest Jaccard similarity: ('Betamethasone (topical)', 'Teprotumumab')
   Jaccard similarity value: 1.0
20. Nodes with highest Jaccard similarity: ('Dexamethasone (topical)', 'Aloe Vera Leaf')
   Jaccard similarity value: 1.0

The Jaccard similarity is a measure used to quantify the similarity between two sets. In the context of your drug network, it is used to measure the similarity between the neighbors of two drugs. A Jaccard similarity value of 1.0 indicates that the two drugs share all their neighbors, suggesting a high level of similarity in their connections within the network.

Here are some interpretations and insights based on the pairs of drugs with the highest Jaccard similarity values:

• Flurbiprofen (ophthalmic), Diclofenac (ophthalmic): Both drugs are nonsteroidal anti-inflammatory drugs (NSAIDs) used in ophthalmic formulations. The Jaccard similarity value of 1.0 implies that they share the same neighbors in the network, suggesting that they may have similar drug interactions and therapeutic uses in eye-related conditions.

• Botulinum toxin type A, Botulinum Toxin Type B: These two drugs are different types of botulinum toxin, which are used in various therapeutic and cosmetic applications. The Jaccard similarity value of 1.0 suggests that they have the same neighbors in the network, implying similar drug interactions and therapeutic applications.

• 3-17. The majority of these pairs involve 'Betamethasone (topical) and other topical corticosteroids. This indicates that these drugs share the same neighbors in the network, suggesting that they have similar drug interactions and therapeutic uses in treating skin conditions, such as eczema, psoriasis, or dermatitis.

• Betamethasone (topical), Chromic chloride: This pair is quite unusual as the two drugs have different therapeutic uses. Betamethasone is a topical corticosteroid, while chromic chloride is a trace element supplement. The Jaccard similarity value of 1.0 suggests that they share the same neighbors in the network, but the reason for this high similarity might require further investigation.

• Betamethasone (topical), Teprotumumab: Teprotumumab is a monoclonal antibody used to treat thyroid eye disease. The high Jaccard similarity with Betamethasone (topical) suggests that they share the same neighbors in the network, which could imply similar drug interactions or overlapping therapeutic uses in eye-related conditions.

• Dexamethasone (topical), Aloe Vera Leaf: Dexamethasone is a topical corticosteroid, while Aloe Vera Leaf is a natural product with various therapeutic uses, including skin conditions. The Jaccard similarity value of 1.0 implies that they share the same neighbors in the network, suggesting that they may have similar drug interactions and therapeutic uses in treating skin conditions.

In conclusion, the high Jaccard similarity values suggest that these drug pairs share the same neighbors in the network, which could imply similar drug interactions or overlapping therapeutic applications. This information could be valuable for understanding drug-drug interactions, drug repositioning, and identifying potential substitutes for specific drugs. However, some unusual pairs might require further investigation to understand the reason behind their high similarity values.

6.6. Link predictions

This algorithm uses the Jaccard coefficient to predict the likelihood of a link forming between two nodes. The Jaccard coefficient is a measure of similarity between two nodes, based on the number of shared neighbors they have in the graph.

The jaccard_coefficient() function calculates the Jaccard coefficient between two nodes, by finding the intersection and union of their sets of neighbors, calculate Jaccard similarity for non-connected node pairs!

The predict_link() function uses the Jaccard coefficient to predict the likelihood of a link forming between two nodes, by simply returning the Jaccard coefficient as the link prediction score.

The main part of the algorithm calculates the link prediction score for all pairs of nodes in the graph that are not currently connected by an edge. It then sorts the scores in descending order and prints out the top 10 pairs of nodes with the highest link prediction scores.

import networkx as nx

# Define the Jaccard similarity function
def jaccard_similarity(u, v, G):
    u_neighbors = set(G.neighbors(u))
    v_neighbors = set(G.neighbors(v))
    intersection = u_neighbors.intersection(v_neighbors)
    union = u_neighbors.union(v_neighbors)
    return len(intersection) / len(union)

# Define a function to calculate Jaccard similarity for non-connected node pairs
def non_connected_jaccard_similarities(G):
    node_pairs = itertools.combinations(G.nodes(), 2)
    similarities = {}
    for u, v in node_pairs:
        if not G.has_edge(u, v):
            similarities[(u, v)] = jaccard_similarity(u, v, G)
    return similarities

# Find all non-connected node pairs
non_connected_node_pairs = [(u, v) for u, v in itertools.combinations(G.nodes(), 2) if not G.has_edge(u, v)]

# Calculate Jaccard similarity for all non-connected node pairs
non_connected_similarities = {pair: jaccard_similarity(pair[0], pair[1], G) for pair in non_connected_node_pairs}

# Sort the similarities in descending order
sorted_non_connected_similarities = sorted(non_connected_similarities.items(), key=lambda x: x[1], reverse=True)

# Print the top 20 pairs of nodes with the highest Jaccard similarity
for i, (pair, similarity) in enumerate(sorted_non_connected_similarities[:20]):
    print(f"{i+1}. Nodes with highest Jaccard similarity (non-connected): {pair}")
    print(f"   Jaccard similarity value: {similarity}")

1. Nodes with highest Jaccard similarity (non-connected): ('Flurbiprofen (ophthalmic)', 'Diclofenac (ophthalmic)')
   Jaccard similarity value: 1.0
2. Nodes with highest Jaccard similarity (non-connected): ('Botulinum toxin type A', 'Botulinum Toxin Type B')
   Jaccard similarity value: 1.0
3. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Hydrocortisone (topical)')
   Jaccard similarity value: 1.0
4. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Triamcinolone (topical)')
   Jaccard similarity value: 1.0
5. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Alclometasone (topical)')
   Jaccard similarity value: 1.0
6. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Amcinonide')
   Jaccard similarity value: 1.0
7. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Clobetasol (topical)')
   Jaccard similarity value: 1.0
8. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Clocortolone (topical)')
   Jaccard similarity value: 1.0
9. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Desonide (topical)')
   Jaccard similarity value: 1.0
10. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Desoximetasone (topical)')
   Jaccard similarity value: 1.0
11. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Diflorasone (topical)')
   Jaccard similarity value: 1.0
12. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Fluocinonide (topical)')
   Jaccard similarity value: 1.0
13. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Flurandrenolide (topical)')
   Jaccard similarity value: 1.0
14. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Halcinonide (topical)')
   Jaccard similarity value: 1.0
15. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Ulobetasol (topical)')
   Jaccard similarity value: 1.0
16. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Methylprednisolone (topical)')
   Jaccard similarity value: 1.0
17. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Prednicarbate (topical)')
   Jaccard similarity value: 1.0
18. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Chromic chloride')
   Jaccard similarity value: 1.0
19. Nodes with highest Jaccard similarity (non-connected): ('Betamethasone (topical)', 'Teprotumumab')
   Jaccard similarity value: 1.0
20. Nodes with highest Jaccard similarity (non-connected): ('Dexamethasone (topical)', 'Aloe Vera Leaf')
   Jaccard similarity value: 1.0

Prediction link of unknown

combined_df

	DDInterID_A	Drug_A	DDInterID_B	Drug_B	Level	Weight
0	DDInter513	Dexamethasone	DDInter582	Dolutegravir	Minor	3
1	DDInter997	Ivacaftor	DDInter582	Dolutegravir	Minor	3
2	DDInter1019	Lamivudine	DDInter1123	Mannitol	Moderate	2
3	DDInter195	Betamethasone	DDInter1961	Zidovudine	Moderate	2
4	DDInter1961	Zidovudine	DDInter246	Budesonide	Moderate	2
...	...	...	...	...	...	...
222341	DDInter856	Heparin	DDInter805	gamma-Aminobutyric acid	Unknown	2
222342	DDInter1683	Sodium bicarbonate	DDInter805	gamma-Aminobutyric acid	Unknown	2
222359	DDInter908	Iloprost	DDInter62	Ambrisentan	Unknown	2
222365	DDInter1951	Warfarin	DDInter1942	Vitamin D	Unknown	2
222368	DDInter1683	Sodium bicarbonate	DDInter1942	Vitamin D	Unknown	2

160235 rows × 6 columns

# Create new weight column with Minor = 3,  Major = 1, Medium = 2 and Unknown = 0
combined_df['Weight_new'] = combined_df['Level'].map({'Minor': 3, 'Major': 1, 'Medium': 2, 'Unknown': 0})

#show unknown values in weight_new
combined_df[combined_df['Weight_new'] == 0].head(30)

	DDInterID_A	Drug_A	DDInterID_B	Drug_B	Level	Weight	Weight_new
23430	DDInter839	Goserelin	DDInter1242	Montelukast	Unknown	2	0.0
23431	DDInter839	Goserelin	DDInter1815	Tiotropium	Unknown	2	0.0
23432	DDInter839	Goserelin	DDInter766	Fluticasone	Unknown	2	0.0
23433	DDInter458	Cyclosporine	DDInter1066	Lidocaine	Unknown	2	0.0
23434	DDInter458	Cyclosporine	DDInter1091	Loratadine	Unknown	2	0.0
23435	DDInter458	Cyclosporine	DDInter1242	Montelukast	Unknown	2	0.0
23436	DDInter458	Cyclosporine	DDInter1332	Olopatadine	Unknown	2	0.0
23437	DDInter458	Cyclosporine	DDInter1426	Pentoxifylline	Unknown	2	0.0
23438	DDInter458	Cyclosporine	DDInter1635	Salmeterol	Unknown	2	0.0
23439	DDInter458	Cyclosporine	DDInter732	Fexofenadine	Unknown	2	0.0
23440	DDInter458	Cyclosporine	DDInter506	Desloratadine	Unknown	2	0.0
23441	DDInter458	Cyclosporine	DDInter1156	Mercaptopurine	Unknown	2	0.0
23442	DDInter458	Cyclosporine	DDInter1533	Promethazine	Unknown	2	0.0
23443	DDInter458	Cyclosporine	DDInter1815	Tiotropium	Unknown	2	0.0
23444	DDInter458	Cyclosporine	DDInter766	Fluticasone	Unknown	2	0.0
23445	DDInter1323	Octreotide	DDInter1242	Montelukast	Unknown	2	0.0
23446	DDInter1323	Octreotide	DDInter766	Fluticasone	Unknown	2	0.0
23447	DDInter451	Cyanocobalamin	DDInter749	Flunisolide	Unknown	2	0.0
23448	DDInter451	Cyanocobalamin	DDInter1066	Lidocaine	Unknown	2	0.0
23449	DDInter451	Cyanocobalamin	DDInter1091	Loratadine	Unknown	2	0.0
23450	DDInter451	Cyanocobalamin	DDInter1242	Montelukast	Unknown	2	0.0
23451	DDInter451	Cyanocobalamin	DDInter1955	Zafirlukast	Unknown	2	0.0
23452	DDInter451	Cyanocobalamin	DDInter1855	Triamcinolone	Unknown	2	0.0
23453	DDInter451	Cyanocobalamin	DDInter885	Hydrocortisone	Unknown	2	0.0
23454	DDInter451	Cyanocobalamin	DDInter1332	Olopatadine	Unknown	2	0.0
23455	DDInter451	Cyanocobalamin	DDInter1426	Pentoxifylline	Unknown	2	0.0
23456	DDInter451	Cyanocobalamin	DDInter1513	Prednisolone	Unknown	2	0.0
23457	DDInter451	Cyanocobalamin	DDInter1012	Ketotifen	Unknown	2	0.0
23458	DDInter451	Cyanocobalamin	DDInter1635	Salmeterol	Unknown	2	0.0
23459	DDInter451	Cyanocobalamin	DDInter732	Fexofenadine	Unknown	2	0.0

6.6. Unknown Link Prediction

We try to predict the link of unknown links.

# Filter edges with 'Unknown' interaction
filtered_edges = combined_df[combined_df['Weight_new'] != 0]

# Create a new graph without the unknown interactions
G_filtered = nx.from_pandas_edgelist(filtered_edges, source='Drug_A', target='Drug_B', edge_attr='Weight_new', create_using=nx.Graph())

# Filter edges with 'Unknown' interaction
unknown_edges = combined_df[combined_df['Weight_new'] != 0]
unknown_node_pairs = [(row['Drug_A'], row['Drug_B']) for _, row in unknown_edges.iterrows()]

unknown_similarities = {pair: jaccard_similarity(pair[0], pair[1], G_filtered) for pair in unknown_node_pairs}

These results show the top 20 pairs of drugs with the highest Jaccard similarity scores among the unknown interactions. A higher Jaccard similarity score indicates that the two drugs share a larger proportion of their neighbors, suggesting that they may have similar interaction profiles.

The interpretation of these results is that the drug pairs with higher Jaccard similarity values are more likely to have similar interaction profiles and, therefore, may potentially interact with each other. In other words, these drug pairs may warrant further investigation to determine if there is indeed a previously unknown interaction between them.

For example, the first pair of drugs with the highest Jaccard similarity score, Etanercept and Golimumab, share 99.43% of their neighbors, indicating that they might have a high likelihood of interacting with each other. Similarly, the second pair, Adalimumab and Certolizumab pegol, share 99.15% of their neighbors, also suggesting a potential interaction.

Keep in mind that these scores should be used as a starting point for further investigation and not as definitive evidence of an interaction. Experimental validation or further research would be required to confirm any potential interactions.

sorted_unknown_similarities = sorted(unknown_similarities.items(), key=lambda x: x[1], reverse=True)

# Print the top 20 pairs of nodes with the highest Jaccard similarity among unknown interactions
for i, (pair, similarity) in enumerate(sorted_unknown_similarities[:20]):
    print(f"{i+1}. Nodes with highest Jaccard similarity (unknown interactions): {pair}")
    print(f"   Jaccard similarity value: {similarity}")

1. Nodes with highest Jaccard similarity (unknown interactions): ('Etanercept', 'Golimumab')
   Jaccard similarity value: 0.9943342776203966
2. Nodes with highest Jaccard similarity (unknown interactions): ('Adalimumab', 'Certolizumab pegol')
   Jaccard similarity value: 0.9915492957746479
3. Nodes with highest Jaccard similarity (unknown interactions): ('Alefacept', 'Rilonacept')
   Jaccard similarity value: 0.9902912621359223
4. Nodes with highest Jaccard similarity (unknown interactions): ('Hyoscyamine', 'Dicyclomine')
   Jaccard similarity value: 0.9900497512437811
5. Nodes with highest Jaccard similarity (unknown interactions): ('Dexbrompheniramine', 'Carbinoxamine')
   Jaccard similarity value: 0.9898305084745763
6. Nodes with highest Jaccard similarity (unknown interactions): ('Mepenzolate', 'Methscopolamine')
   Jaccard similarity value: 0.9896373056994818
7. Nodes with highest Jaccard similarity (unknown interactions): ('Enoxaparin', 'Dalteparin')
   Jaccard similarity value: 0.9885057471264368
8. Nodes with highest Jaccard similarity (unknown interactions): ('Anakinra', 'Alefacept')
   Jaccard similarity value: 0.9870550161812298
9. Nodes with highest Jaccard similarity (unknown interactions): ('Alefacept', 'Canakinumab')
   Jaccard similarity value: 0.987012987012987
10. Nodes with highest Jaccard similarity (unknown interactions): ('Anakinra', 'Canakinumab')
   Jaccard similarity value: 0.987012987012987
11. Nodes with highest Jaccard similarity (unknown interactions): ('Etanercept', 'Adalimumab')
   Jaccard similarity value: 0.9859154929577465
12. Nodes with highest Jaccard similarity (unknown interactions): ('Adalimumab', 'Golimumab')
   Jaccard similarity value: 0.9859154929577465
13. Nodes with highest Jaccard similarity (unknown interactions): ('Daunorubicin', 'Idarubicin')
   Jaccard similarity value: 0.9845758354755784
14. Nodes with highest Jaccard similarity (unknown interactions): ('Clidinium', 'Mepenzolate')
   Jaccard similarity value: 0.9844559585492227
15. Nodes with highest Jaccard similarity (unknown interactions): ('Clidinium', 'Methscopolamine')
   Jaccard similarity value: 0.9844559585492227
16. Nodes with highest Jaccard similarity (unknown interactions): ('Orciprenaline', 'Formoterol')
   Jaccard similarity value: 0.983957219251337
17. Nodes with highest Jaccard similarity (unknown interactions): ('Anakinra', 'Rilonacept')
   Jaccard similarity value: 0.9838709677419355
18. Nodes with highest Jaccard similarity (unknown interactions): ('Canakinumab', 'Rilonacept')
   Jaccard similarity value: 0.9838187702265372
19. Nodes with highest Jaccard similarity (unknown interactions): ('Clemastine', 'Carbinoxamine')
   Jaccard similarity value: 0.9831649831649831
20. Nodes with highest Jaccard similarity (unknown interactions): ('Etanercept', 'Certolizumab pegol')
   Jaccard similarity value: 0.9831460674157303